remote sensing

Article
Implementation of the Burned Area Component of the
Copernicus Climate Change Service: From MODIS to
OLCI Data
Joshua Lizundia-Loiola 1, * , Magí Franquesa 1 , Martin Boettcher 2 , Grit Kirches 2 , M. Lucrecia Pettinari 1
and Emilio Chuvieco 1

1 Department of Geology, Geography and the Environment, Universidad de Alcalá,

Environmental Remote Sensing Research Group, Calle Colegios 2, 28801 Alcalá de Henares, Spain;
[email protected] (M.F.); [email protected] (M.L.P.); [email protected] (E.C.)
2 Brockmann Consult GmbH, Chrysanderstr 1, 21029 Hamburg, Germany;
[email protected] (M.B.); [email protected] (G.K.)
* Correspondence: [email protected]

Abstract: This article presents the burned area (BA) product of the Copernicus Climate Change
Service (C3S) of the European Commission. This product, named C3SBA10, is based on the adaptation
to Sentinel-3 OLCI images of a BA algorithm developed within the Fire Climate Change Initiative
(FireCCI) project, which used MODIS data. We first reviewed the adaptation process and then
analysed the results of both products for common years (2017–2019). Comparisons were performed
using four different grid sizes (0.05◦ , 0.10◦ , 0.25◦ , and 0.50◦ ). Annual correlations between the





 two products ranged from 0.94 to 0.99. Global BA estimates were found to be more similar when
the two Sentinel-3 satellites were active (2019), as the temporal resolution was closer to that of the
Citation: Lizundia-Loiola, J.;
MODIS sensor. Global validation was performed using reference data derived from Landsat-8 images,
Franquesa, M.; Boettcher, M.;
Kirches, G.; Pettinari, M.L.;
following a stratified random sampling design. The C3SBA10 showed commission errors between
Chuvieco, E. Implementation of the 16 and 21% and omission errors from 48 to 50%, similar to those found in the FireCCI product. The
Burned Area Component of the temporal reporting accuracy was also validated using 19 million active fires. In total, 87% of the
Copernicus Climate Change Service: detections were made within 10 days after the fire by both products. The high consistency between
From MODIS to OLCI Data. Remote both products ensures global BA data provision from 2001 to the present. The datasets are freely
Sens. 2021, 13, 4295. https://doi.org/ available through the Copernicus Climate Data Store (CDS) repository.
10.3390/rs13214295

Keywords: Sentinel-3; OLCI; MODIS; burned area; Copernicus; Climate Change Service; FireCCI; ESA
Academic Editor: Leonor Calvo

Received: 20 September 2021

Accepted: 21 October 2021
1. Introduction
Published: 26 October 2021
Interactions between climate and fire are bidirectional. On the one hand, climate
Publisher’s Note: MDPI stays neutral
impacts fire regimes [1–4], mostly by modifying temperature and precipitation patterns,
with regard to jurisdictional claims in
which in turn impacts fire ignition and behaviour through changes in soil and fuel moisture,
published maps and institutional affil- vegetation productivity, and fuel availability [5–8]. On the other hand, biomass burning is
iations. a critical source of aerosols and greenhouse gases that directly affect atmospheric chem-
istry [9,10], carbon budgets [11,12], and carbon stocks [13]. Emissions from fires increase
aerosol optical depth, modifying the radiation budget and, thus, warming the lower
atmosphere, which affects regional temperature, clouds, and precipitation patterns [14,15].
Copyright: © 2021 by the authors.
Fires have important ecological implications as well. They are closely related to biodi-
Licensee MDPI, Basel, Switzerland.
versity, either favouring or degrading it depending on fire severity and persistency [16–18].
This article is an open access article
Furthermore, wildfires have significant societal impacts, particularly when extreme events
distributed under the terms and occur as a result of climate anomalies. Large damage to people’s lives, health and in-
conditions of the Creative Commons frastructure have been observed in recent catastrophic fire seasons of Southern Europe,
Attribution (CC BY) license (https:// Australia and Western USA [2,8,19].
creativecommons.org/licenses/by/ For those reasons, Fire Disturbance was included by the Global Climate Observing
4.0/). System (GCOS) as one of the Essential Climate Variables (ECV) [20], acknowledging

Remote Sens. 2021, 13, 4295. https://doi.org/10.3390/rs13214295 https://www.mdpi.com/journal/remotesensing

Remote Sens. 2021, 13, 4295 2 of 25

its relevance to characterising the Earth’s climate system. The European Space Agency
(ESA) responded to the GCOS demands by promoting the Climate Change Initiative (CCI)
programme in 2009. Its main objective was to develop long-term datasets of ECVs based
on satellite observations. This programme was initially focused on 13 ECVs (extended to
26 ECVs in 2018), each of them addressed within a dedicated project that encompassed
algorithm development, validation, uncertainty characterisation, and large-scale earth
observations data processing, among other tasks [21].
Fire disturbance was one of the initial ECVs tackled within the CCI programme
(FireCCI project, https://climate.esa.int/en/projects/fire/, accessed on 21 October 2021),
aiming to produce long-term time series of global burned area (BA) data. Although
BA information is required for a wide range of applications [22,23], the FireCCI project
was mainly oriented towards climate modellers and, hence, BA data have been mainly
used for the characterisation of fire emissions [9,24] and the parameterisation of Dynamic
Global Vegetation Models (DGVMs) [25–28]. These applications strongly benefit from an
extended temporal coverage of BA datasets, while assuring their temporal and spatial
consistency [29]. The most recent BA products developed within the FireCCI project
include the FireCCI51, based on MODIS data, which extends from 2001 to 2019 [30], and
the FireCCILT11 (1982–2018), based on AVHRR-LTDR data [31].
A few years after the ESA CCI programme started, the European Commission launched
the Copernicus Climate Change Service (C3S, https://climate.copernicus.eu/, accessed
on 21 October 2021), which now includes more than 20 ECVs [32]. This service builds
upon the research and development carried out by the scientific community, especially the
one associated to the CCI programme, and it is the service in charge of the operational
production of those ECV datasets. Regarding BA, the FireCCI51 product was transferred
to the C3S for the historical record, while demanding the generation of a new BA dataset
based on European Copernicus satellites. The new product should guarantee consistency
with the FireCCI51 data, as this dataset was developed only for research purposes, and
it would no longer be processed. Adapting both the BA algorithm and the product to
the new Copernicus missions would assure an operational continuity of the BA products.
Since FireCCI51 was generated from the 250 m resolution near-infrared (NIR) band of the
MODIS sensor [30], it was decided to base the continuity of this BA dataset on the Ocean
and Land Colour Instrument (OLCI) on board Sentinel-3 (S3), which has a similar spatial
and temporal resolution.
The adaptation of algorithms and products to different sensors is often performed
when long-term time series are required, since the lifetime of single missions is generally
limited. Coupling datasets from different input sensors extends the length of the series or
helps fill existing gaps. The change from one input data source to another is not automatic,
as each sensor has its particular characteristics in terms of spatial, temporal and radiometric
resolution, which determine the performance of the algorithm. For this reason, a thorough
review of the algorithm is required, in order to adapt it to the characteristics of the new
sensor. For instance, the SeaWIFS Ocean Aerosol Retrieval (SOAR) algorithm, originally
designed for the retrieval of aerosol optical depth (AOD) over water from the SeaWIFS
sensor [33], and the Deep Blue algorithm to retrieve AOD over land [34], have been adapted
to the Suomi-NPP Visible Infrared Imaging Radiometer Suite (VIIRS) sensor [35,36] and
to the AVHRR sensor [37], requiring several algorithm adaptations and optimisations.
Another example is the sea surface temperature (SST) dataset of C3S (https://cds.climate.
copernicus.eu/cdsapp#!/dataset/satellite-sea-surface-temperature, accessed on 21 October
2021), which derives independent SST Level 3 products from AVHRR, ATSR1, ATSR2,
AATSR and SLSTR sensors [38]. To assure the consistency of the different products, the
harmonisation of the sensor input data [39] and the adaptation of some parts of the SST
retrieval algorithm [40] were performed. Although a Level 4 multi-sensor product exists,
which provides spatially complete information, the authors indicate that the process of
interpolation for gap-filling entails that feature resolution is degraded relative to the lower-
Remote Sens. 2021, 13, 4295 3 of 25

level data [41], and for this reason, different products from the individual sensors are still
provided, and are useful.
This paper presents the adaptation of the FireCCI51 BA algorithm to the OLCI sensor
and the generation of the new BA dataset, named C3SBA10, which is now part of the
C3S service. This product is the first global BA product derived from OLCI data. This
manuscript briefly describes the original algorithm (FireCCI51) in Section 2.1, while the
pre-processing, adaptation, uncertainty characterisation, generation, and distribution of
the OLCI-based BA dataset are explained in Sections 2.2–2.5. The C3SBA10 data were
produced from 2017 to the present. This dataset was designed to be as consistent as
possible with its predecessor FireCCI51 (2001–2019) to ensure that multi-decadal analyses
can benefit from both datasets uninterruptedly. For this reason, this paper emphasises
the inter-comparison analysis between FireCCI51 and C3SBA10 datasets for the common
period (2017–2019), pointing out similarities and discrepancies (Section 2.6). The products
were spatially validated using reference data obtained from multitemporal Landsat-8
images, and temporally validated using active fire information (Section 2.7). The results of
the BA detections for the common time series (2017–2019) as well as the quality assessment
of the product are presented in Section 3, followed by the discussion and conclusion
(Sections 4 and 5, respectively).

2. Materials and Methods

2.1. FireCCI51 Algorithm
The FireCCI51 algorithm was developed to take advantage of the higher spatial
resolution of the Red/NIR bands of MODIS (250 m). It follows a hybrid approach that
combines active fires and reflectance changes caused by burnings (see Figure 1, with
further details in [30]). The algorithm first created monthly composites of NIR reflectance
by searching for the minimum NIR daily surface reflectance within the monthly period
while maximising the proximity to the date determined by the nearest active fire. Since
NIR typically decreases after a fire event, a second monthly variable that retains the relative
NIR drop from the previous composite was computed as well. To avoid problems related
to cloud and cloud shadows, pixel state information, which indicates the quality of the
observation, was used to filter out those pixels. The third variable was generated from
neighbour active fires that are sufficiently close in space (<1875 m) and time (<4 days) to
be aggregated into spatio-temporal clusters (STC). To improve computational efficiency, a
fuel mask (burnable categories) was derived from land cover data. It was used to remove
urban areas, bare areas, water bodies, and permanent snow and ice classes from further
processing [30].
The detection of burned pixels was performed following a two-phase approach.
The first step aimed to minimise commission errors by selecting seed-pixels with a high
probability of being burned (seed-phase). NIR and relative NIR drop thresholds were
obtained for each STC based on specific burned and unburned samples. The burned sample
was defined by the active fires that belong to the STC. The unburned one was composed of
pixels from the vicinity of the STC, i.e., those located within a strip between 10 and 20 km
from the STC. Using in each case the thresholds of the corresponding STC, active fires
were filtered out to obtain seed-pixels. The second phase of the algorithm proceeded with
iterative contextual growth around each seed-pixel, the stop criteria being the thresholds of
the STC to which the seed-pixel belonged (growing-phase). The purpose of this step was
to minimise omission errors by better shaping the burned patch.
The final result was a monthly classification, wherein each burned pixel was labelled
by its day of detection, i.e., the day of the year from which the NIR value of the monthly
composite was extracted.
 Remote Sens. 2021, 13, x FOR PEER REVIEW 4 of 26

Remote Sens. 2021, 13, 4295 The final result was a monthly classification, wherein each burned pixel was labelled
4 of 25
by its day of detection, i.e., the day of the year from which the NIR value of the monthly
composite was extracted.

Figure1.1.Simplified
Figure Simplifiedversion
versionof
ofthe
themain
mainscheme
schemeof
ofthe
theglobal
globalBA
BAalgorithm.
algorithm.Adapted
Adaptedfrom
from[30].
[30].

2.2.
2.2.Input
InputData
Datafor forthe
theC3SBA10
C3SBA10ProductProduct
2.2.1. OLCI Surface Directional Reflectances
2.2.1. OLCI Surface Directional Reflectances
The S3 satellites were designed to provide operational ocean and land observation
The S3 satellites were designed to provide operational ocean and land observation
services (https://sentinel.esa.int/web/sentinel/missions/sentinel-3, accessed on 21 Octo-
services (https://sentinel.esa.int/web/sentinel/missions/sentinel-3, accessed on 21 October
ber 2021). They represent the continuity of the European Remote Sensing satellites (ERS-1
2021). They represent the continuity of the European Remote Sensing satellites (ERS-1
and ERS-2), the Envisat, and the Satellite Pour l’Observation de la Terre (SPOT). S3s have a
and ERS-2), the Envisat, and the Satellite Pour l’Observation de la Terre (SPOT). S3s have
low-altitude (814.5 km), high-inclination (98.65◦ ) sun-synchronous Earth orbit and need
a low-altitude (814.5 km), high-inclination (98.65°) sun-synchronous Earth orbit and need
27 days to complete a full cycle (385 orbits). The first satellite, called S3A, was launched in
27 days to complete a full cycle (385 orbits). The first satellite, called S3A, was launched in
February 2016 and the second one, called S3B, was launched two years later. The orbit of
February
S3B 2016 and
is identical thebut
to S3A second
it fliesone, called
±140 S3B,
◦ out was launched
of phase to improve tworevisit
years time.
later. The
The design
orbit of
S3B is identical to S3A but it flies ±140° out of phase to improve revisit
life of S3 is 7.5 years (consumables for 12 years), but the mission is expected to last longer, time. The design
life of S3 is 7.5 years (consumables
with two additional satellites already funded. for 12 years), but the mission is expected to last longer,
withThe twoOLCI
additional
sensor,satellites
one of S3’salready
mainfunded.
payloads, is a push-broom imaging spectrometer
The OLCI sensor, one of S3’s
composed of five cameras that are tiled 12.6 main payloads,
◦ in the is a push-broom
western directionimaging
to mitigatespectrometer
potential
composed of five cameras that are tiled 12.6° in the western direction
sun-glint effects (https://sentinel.esa.int/web/sentinel/technical-guides/sentinel-3-olci, to mitigate potential
sun-glint effects (https://sentinel.esa.int/web/sentinel/technical-guides/sentinel-3-olci,
accessed on 21 October 2021). It has a swath width of 1270 km (Field-of-View = 68.6◦ )
accessed
and on 21
it offers October
global 2021).atIt 300
coverage has ma swath
everywidth
~2 or of
~11270
day km (Field-of-View
depending on the =number
68.6°) and of
satellites available (S3A or S3A + S3B, respectively), the latitude, and the applicationof
it offers global coverage at 300 m every ~2 or ~1 day depending on the number satel-
(ocean
lites
or available
land). The OLCI(S3A instrument
or S3A + S3B, respectively),
includes the latitude,
21 spectral and the application
bands (400–1020 nm), 15 taken (ocean
from or
land).
the The OLCI
precursor instrument
sensor MERIS includes
on board21 thespectral
ENVISAT bands (400–1020
satellite, andnm), 15 taken
6 extra fromthat
channels the
precursor
were sensor
included MERIS on
to improve theboard the ENVISAT
atmospheric satellite,
and aerosol and 6 extra
corrections channels
(Table that were
1). OLCI was
included to improve the atmospheric and aerosol corrections
initially designed for ocean monitoring, but it has been successfully used in several (Table 1). OLCI was ini-
land
tially designed
applications for ocean
[42,43]. MERIS monitoring,
data were the but basis
it hasofbeen successfully
the first global BAused in several
dataset within land
the
applications
FireCCI [42,43].
[44], and MERISitdata
therefore, was were the basis
expected of thewould
that OLCI first global
showBA dataset
similar BA within
detection the
FireCCI [44],although
capabilities, and therefore, it wastemporal
with better expectedresolution
that OLCI(3-day
wouldrevisit
show time
similarforBA detection
MERIS and
1capabilities,
day for OLCI, although
when thewithtwobetter temporal
S3s are resolution (3-day revisit time for MERIS and 1
available).
day Thefor OLCI, when1the
OLCI Level two S3s
product are available).
(https://sentinel.esa.int/web/sentinel/user-guides/sentinel-
The
3-olci/product-types/level-1b, OLCI accessed on 21 Level
October 2021) was 1 converted to surfaceproduct
(https://sentinel.esa.int/web/sentinel/user-guides/sentinel-3-olci/product-types/level-1b,
directional reflectance (SDR) using an automated pre-processing chain that generated
OLCI
accessed Level
on3 21SDR. The first
October module
2021) of the chain
was converted toconverted the top-of-atmosphere
surface directional reflectance (SDR)spectralus-
radiance to the apparent SDR. Then, a pixel identification module calculated a set of pixel
classification attributes, such as clear, cloud, snow/ice, cloud shadows, etc. The third
module included an atmospheric correction algorithm, considering the absorbing and
Remote Sens. 2021, 13, 4295 5 of 25

scattering effects of atmospheric gases, in particular ozone, oxygen and water vapour,
of the scattering of air molecules (Rayleigh scattering) and the correction of absorption
and scattering due to aerosol particles. The atmospheric correction was first developed
in the GlobAlbedo project [45]. The last processing module of the pre-processing chain
(compositing and mosaicking) was applied to retrieve the final Level 3 products. These
1-day SDR composites were derived from an input set of single satellite observations (i.e.,
SDR and pixel classification data described previously). Thus, those single observations
were (i) reprojected onto a Plate Carrée grid, (ii) temporally aggregated for given binning
cells (tiles), and (iii) mosaicked based on the binning cells onto a Level 3 product of
10 × 10 degrees tiles (Figure 2).

Table 1. OLCI bands characteristics. The band indicated in grey was the one used as input for the
C3SBA10 algorithm.

Band Band Centre (nm) Bandwidth (nm) MERIS Heritage

Oa01 400 15 No
Oa02 412.5 10 Yes
Oa03 442.5 10 Yes
Oa04 490 10 Yes
Oa05 510 10 Yes
Oa06 560 10 Yes
Oa07 620 10 Yes
Oa08 665 10 Yes
Oa09 673.75 7.5 No
Oa10 681.25 7.5 Yes
Oa11 708.75 10 Yes
Oa12 753.75 7.5 Yes
Oa13 761.25 2.5 Yes
Oa14 764.375 3.75 No
Oa15 767.5 2.5 No
Oa16 778.75 15 Yes
Oa17 865 20 Yes
Oa18 885 10 Yes
Oa19 900 10 Yes
Oa20 940 20 No
Oa21 1020 40 No

2.2.2. Auxiliary Data

Information on active fires was used to guide both the seed and the growing phase
of the BA algorithm. The input product was the MODIS MCD14ML Collection 6, which
provides global monthly information on the location of thermal anomalies using both Terra
and Aqua MODIS thermal bands [46] at 1 km spatial resolution. The MCD14ML product is
provided in the ASCII format, and one of its attributes specifies the presumed origin of
the thermal anomaly: 0 for presumed vegetation fire, 1 for active volcano, 2 for other static
land sources, and 3 for offshore thermal anomalies. Among these categories, only those
anomalies labelled as 0 were considered.
Global, annual land cover (LC) maps, which were provided by the Land Cover C3S (https:
//cds.climate.copernicus.eu/cdsapp#!/dataset/satellite-land-cover?tab=overview, accessed
on 21 October 2021), were used as well. This LC product was brokered to C3S from the CCI
programme and extended for the years 2016–2019 using data at 333 m spatial resolution
from PROBA-V. This satellite, which was launched in 2013, was designed to bridge the
gap in vegetation remote sensing between SPOT-VGT (1998–2014) and the S3 [47]. Land
cover was used by the algorithm to mask out unburnable areas, i.e., bare areas, urban areas,
water bodies, and permanent snow and ice classes, to decrease the amount of data to be
processed, and to report which land cover classes were burned.
 accessed on 21 October 2021), were used as well. This LC product was brokered to C3S
from the CCI programme and extended for the years 2016–2019 using data at 333 m spa-
tial resolution from PROBA-V. This satellite, which was launched in 2013, was designed
to bridge the gap in vegetation remote sensing between SPOT-VGT (1998–2014) and the
Remote Sens. 2021, 13, 4295
S3 [47]. Land cover was used by the algorithm to mask out unburnable areas, i.e.,6 of
bare
25
areas, urban areas, water bodies, and permanent snow and ice classes, to decrease the
amount of data to be processed, and to report which land cover classes were burned.

Figure2.2.The
Figure The1010××10-degree
10-degreetiles
tilesused
usedforfor
thethe processing
processing of the
of the OLCI OLCI global
global BA product.
BA product. A of
A total total
273oftiles
273were
tilesprocessed,
were pro-
cessed,
of whichof13which 13 were
were used used to
to adapt theadapt the algorithm
algorithm (red and(red
greenand green tiles).
tiles).

2.3. C3SBA10
2.3. C3SBA10 Product
Product Algorithm
Algorithm
The adaptation
The adaptationofofthe the FireCCI51
FireCCI51 algorithm
algorithm to theto OLCI
the OLCIsensor sensor was based
was based on theon the
study
studyshown
sites sites shown
in Figure in 2,Figure
as they 2, encompassed
as they encompassed a wide varietya wideof variety of vegetation
vegetation types andtypes fire
and fire regimes.
regimes. These study These study
sites weresites
the were
thirteentheequivalent 10◦ × 10◦10°
thirteen equivalent OLCI× 10° OLCI
tiles of thetiles of the
MODIS
MODIS standard,
standard, sinusoidalsinusoidal
tiles used tiles
in theused in the development
development of the original of the originalalgorithm
FireCCI51 FireCCI51 al-
[30].
gorithm
Six [30].
of those tilesSixwere
of those tiles were
distributed distributed
in regions whereinfire regions wherefrom
perimeters fire different
perimeters from
official
differentwere
services official servicesThe
available. were available.
western coastThe western
of the Unitedcoast of the
States, United States,
in California, in Cali-
was chosen
as representative
fornia, was chosen of temperate forests. Fire
as representative perimetersforests.
of temperate for thisFire
areaperimeters
were obtained through
for this area
the
wereFireobtained
and Resource Assessment
through the Fire Programand (FRAP,Resource http://frap.fire.ca.gov/,
Assessment Program accessed (FRAP,on
21 October 2021). Northern
http://frap.fire.ca.gov/, Australia
accessed on 21was Octoberselected
2021). as Northern
representative of tropical
Australia savanna.
was selected as
Fire perimeters were downloaded from the Northern Australian
representative of tropical savanna. Fire perimeters were downloaded from the Northern Fire Information (NAFI,
http://www.firenorth.org.au/nafi3/,
Australian Fire Information (NAFI, accessed on 21 October 2021). Finally,accessed
http://www.firenorth.org.au/nafi3/, as an exampleon 21
of the boreal
October 2021).forests, fireas
Finally, perimeters
an example wereof downloaded
the boreal forests, from thefire Canadian
perimetersNational
were down- Fire
Database
loaded (CNFDB, from http://cwfis.cfs.nrcan.gc.ca/ha/nfdb,
the Canadian National accessed
Fire on 21 October
Database 2021).
(CNFDB,
The remaining seven tiles were used to
http://cwfis.cfs.nrcan.gc.ca/ha/nfdb, visuallyon
accessed check that no major
21 October 2021). problems
The remainingarose whenseven
applying the adapted algorithm to the new sensor, e.g.,
tiles were used to visually check that no major problems arose when applying the border effects in Central Africa,
problems on the thresholds
adapted algorithm to the new duesensor,
to contrasted fire regimes
e.g., border effects within
in Centralthe same
Africa,tile in Angola,
problems on
or
the thresholds due to contrasted fire regimes within the same tile in Angola, orThe
region-growing problems due to high BA/active fire ratios found near Kazakhstan. re-
adaptation
gion-growing of the algorithm
problems duetoto OLCI
highand its integration
BA/active fire ratiosinto found
the C3Snearsystem were carried
Kazakhstan. The
out during 2018
adaptation of theand, hence, the
algorithm training
to OLCI anddataset’s temporal
its integration intocoverage
the C3S wassystemlimited
weretocarried
2017,
when only S3A
out during 2018was and,available.
hence, the training dataset’s temporal coverage was limited to 2017,
Most of
when only S3A wasthe parameters
available. that control the FireCCI51 BA algorithm were independent
of theMost
base ofsensor
the parameters thatBA,
used to detect suchthe
control as the time-gap
FireCCI51 BAused to temporally
algorithm aggregate
were independent
active
of the base sensor used to detect BA, such as the time-gap used to temporally However,
fires or the absolute thresholds that were fixed for global detection [30]. aggregate
parameters
active fires orlinked to a distance
the absolute mightthat
thresholds be affected
were fixed byfortheglobal
spatialdetection
resolution ofHowever,
[30]. both the
surface reflectance product and the active fire product. The fixed distance used to spatially
parameters linked to a distance might be affected by the spatial resolution of both the
aggregate active fires into the same cluster (Section 2.1), for example, might depend on the
spatial resolution of the input active fire product. Higher-spatial resolution thermal bands
can lead to higher densities of active fire pixels within the same burned patch [48,49], and
therefore, the distance that is needed to spatially aggregate them could be smaller. In this
case, there was no difference between the input active fire product used in the development
of the original algorithm and the adapted one (Section 2.2.2). Similarly, the distances used
to define the unburned region around active fire clusters (Section 2.1) depend on both the
spatial resolution of the input surface reflectance and the density of active fires per burned
Remote Sens. 2021, 13, 4295 7 of 25

patch. The spatial resolution of the base sensor directly affects the number of surface
reflectance pixels that fall into the unburned strip, and hence could be used to estimate the
thresholds. The same distance could imply the selection of much more pixels when the
spatial resolution of the input sensor was high, while coarser resolutions could lead to a
number of pixels that may be insufficient. The spatial resolutions of the input sensors used
in the original algorithm (250 m) and the adapted one (300 m) were considered sufficiently
similar to retain the parameters that defined the unburned strip.
Since the FireCCI51 BA algorithm worked with a single NIR band of the MODIS
sensor (841–876 nm), the first step for adapting the algorithm was to find the most suitable
OLCI band (Table 1). Based on a sensitivity analysis of the MERIS bands carried out by [50]
during the development of the first global BA dataset of the FireCCI project (FireCCI41 [44]),
two NIR bands were considered candidates to be the input of the adapted BA algorithm:
Oa12 centred at 753.75 nm and Oa17 centred at 865 nm (Table 1). These spectral bands
were the most sensitive for BA detection, although no significant differences were found
between them [50]. Therefore, the adapted BA algorithm was run twice (one per band)
using the training dataset to verify the performance of both bands. The assessment against
the services’ fire perimeters showed slightly higher accuracy measures of the 865 nm band
over the 753.75 nm one (same commission error and ~1% lower omission error). None of
the bands showed unexpected anomalies in the rest of the tiles. Taking into account these
results, we finally selected the 865 ± 10 nm OLCI band, which also had maximum spectral
similarity with the MODIS band used for the FireCCI51 algorithm.

2.4. Uncertainty Characterisation

Uncertainty characterisation is increasingly being demanded by the end-users of BA
products, since it helps in the parameterisation of climate models [29]. Although important
advances have been made in the last years regarding the uncertainty characterisation of
ECVs [51–53], methods to standardise the generation of this variable are still needed [22].
Therefore, the CS3BA10 algorithm follows the same approach used in the FireCCI51
product. Four input variables related to different phases of the original algorithm were
used to compute the estimation of BA detection uncertainty (pb): monthly NIR composite
(NIR), monthly relative NIR drop (Rel∆NIR), number of valid observations in the first
10 post-fire days (obs) and the distance to the nearest seed-pixel (dist). The uncertainty
was computed using a logistic regression. Based on those four variables and the results
from 2008 of the study sites that were used to develop the original algorithm, the final
coefficients of the regression were obtained.
In the case of the adapted C3SBA10 algorithm, the coefficients of the logistic regression
were updated (Equation (1)) using the four input variables and results of the training dataset
described in Section 2.3:
1
pb = , where
1 + ec (1)
c = −(4.068 − 0.002926· N IR + 0.003942· Rel∆N IR − 0.01303·obs − 17.29·dist)

2.5. Product Generation

The adapted algorithm was processed on a monthly basis for each 10◦ × 10◦ tile, and
hence the raw outputs were provided per month and tile. The pre-processing and BA
processing steps took advantage of the independence of tiles, and were produced in parallel
for multiple tiles. The raw outputs were formatted, mainly following the specifications of
the FireCCI project [54], to obtain the pixel and grid products that were delivered to the final
users in the NetCDF4 format. The pixel products include the date of the detected burned
pixel at the original spatial resolution (300 m) distributed in six continental tiles (Figure 3),
plus the uncertainty and land cover of those pixels. The grid products include aggregated
BA at 0.25◦ spatial resolution, burned land cover, uncertainty, fraction of burnable area,
and fraction of observed area (Table 2).
 parallel for multiple tiles. The raw outputs were formatted, mainly following the speci-
fications of the FireCCI project [54], to obtain the pixel and grid products that were de-
livered to the final users in the NetCDF4 format. The pixel products include the date of
the detected burned pixel at the original spatial resolution (300 m) distributed in six con-
Remote Sens. 2021, 13, 4295 tinental tiles (Figure 3), plus the uncertainty and land cover of those pixels. The8grid of 25
products include aggregated BA at 0.25° spatial resolution, burned land cover, uncer-
tainty, fraction of burnable area, and fraction of observed area (Table 2).

Figure
Figure 3.
3. Continental
Continental tiles
tiles of
of the
the global
global BA
BA pixel
pixel product.
product.

Table 2. Pixel
Tableand grid products’
2. Pixel specifications.
and grid products’ specifications.
Spatial Reso-
Product Spatial Resolution
Product Layers Layers Description
Description
lution
Julian day or day of the year when the burned pixel
Pixel 300 m JD Julian day or day of the year when
was detected
Pixel 300 m JD
Confidence level of the burned
the pixel
classified was
pixel detected
(both burned
CL
Confidence
and unburned)level of the classified
LC CL Land cover
pixel class
(boththat was burned
burned and unburned)
Grid 0.25◦ Burned area Sum of the burned area within the grid cell
LC Land cover class that was burned
Standard error Estimation of the standard error of the burned area
Sum of the
Fraction of the grid cell thatburned area(vegetated
could burn within the
Grid 0.25°
Fraction of burnable areaBurned area
land covers)grid cell
Fraction of the burnable area
Estimation of thatstandard
the was observed
error of
Fraction of observed area
Standard error during the month
the burned area
Burned area of each land Sum of the burned area within the grid cell per land
cover class Fraction
coverofclass
the grid cell that could
Fraction of burnable area
burn (vegetated land covers)
Fraction of the burnable area that
Fraction
The processing was run on 273oftiles
observed area
that encompass the main burnable areas of the
was observed during the month
Earth, as determined by the input land cover data. The C3SBA10 extended the spatial
Burned area of each land Sum of the burned area within the
coverage provided by the FireCCI51 dataset, including northern latitudes between 70◦ and
cover
80◦ of North America and Greenland, class
since grid
rare fire events cellobserved
were per land in
cover class
recent years
in these regions [55,56]. This change affected the original bounding box of Area 1 of the
pixelThe processing
product definedwas run within
by [54] on 273the
tiles that encompass
FireCCI, the main
whose eastern burnable
border areas offrom
was extended the
Earth,
◦ as determined
◦ by the input
50 W to 26 W in C3SBA10 (Figure 3). land cover data. The C3SBA10 extended the spatial
coverage provideddataset
The C3SBA10 by the has
FireCCI51 dataset,processed
been regularly includingsince
northern latitudes
January betweenup70°
2017, ensuring to
three years of overlap (2017–2019) with FireCCI51.

2.6. Intercomparison between FireCCI51 and C3SBA10

Consistency between products is of prime interest for long-term analyses, which is
a critical component of climate modelling [22,29]. From this perspective, since C3SBA10
was conceived as a continuation of FireCCI51, several analyses were performed to assess
the similarities and differences between both datasets. This agreement may be influenced
either by the differences between the main input reflectance datasets used by each of them
(865 ± 10 nm for C3SBA10 and 858.5 ± 17.5 nm for FireCCI51), by the changes in the
spatial resolution (300 m for C3SBA10 and 250 m for FireCCI51), or by the atmospheric and
geometric correction processes applied to both sensors (OLCI for C3SBA10 and MODIS
for FireCCI51).
Remote Sens. 2021, 13, 4295 9 of 25

A fundamental step when comparing spatial and temporal trends of global BA datasets
is the definition of the grid size to aggregate the BA information. Although different spatial
and temporal comparison units have been previously used [57–60], we were guided by
previous intercomparison efforts performed by the climate community. Within this scope,
one notable example is the “fire model intercomparison project” (FireMIP), which system-
atically compared different fire-enabled DGVMs using global monthly BA data aggregated
at the 0.5◦ resolution [27]. Statistical analyses that target the identification of factors that
control fire patterns at global and regional scales commonly require monthly BA data
aggregated at 0.25◦ or coarser spatial resolutions [5,26,61,62]. However, several global fire
emission databases are available at 0.1◦ (e.g., [63] updated) and even at 0.05◦ (e.g., [64])
following the increasing spatial resolution of some global atmospheric chemistry mod-
els [65,66]. Thus, it was decided to aggregate BA information into four different geographic
grid cell sizes, i.e., 0.05◦ , 0.1◦ , 0.25◦ , and 0.5◦ , on a monthly and annual basis. Since both
FireCCI51 and C3SBA10 are distributed on geographic projection, the effect of the latitude
on pixel size was corrected using a cosine weighting factor (earth radius = 6,378,137.0 m)
when aggregating the BA.
The results were also aggregated spatially into eight biomes, following the same classifica-
tion of the validation exercise. Four comparison cases were identified: biome-specific monthly
BA, biome-specific annual BA, global monthly BA and global annual BA. In total, 1404 scatter-
plots (3 years × (12 monthly + 1 annual) × (eight biomes + 1 global) × 4 spatial resolutions)
were generated based on these comparison cases and the four comparison grids defined above
for the analysed years (2017–2019). Three metrics were used to determine the agreement
of the products for a given case. A total least squares (TLS) regression was calculated for
each scatter plot to obtain the slope and bias between products. TLS was selected, since it
does not assume the dependency of the variables [59]. In addition, Pearson’s correlation
and the root mean square error (RMSE) were used to estimate the agreement between the
products [60].

2.7. Validation
2.7.1. Spatial Assessment
A spatial accuracy assessment was conducted for the period 2017–2019 for both
the FireCCI51 and C3SBA10 products. An independent reference BA dataset derived
from Landsat-8 OLI imagery was produced in compliance with the CEOS LPVS stage
3 validation requirements [67], which implies a rigorous statistical selection of reference
sites representing diverse global conditions. Hence, for each calendar year, 100 sample
units were randomly selected using a stratified random sampling design. The sampling
units were spatially defined by the Thiessen scene areas (TSAs) constructed by [68] and [69]
from the Landsat-frames World Reference System 2 (WRS-2). TSAs enable the partition
of the Earth’s surface into non-overlapping spatial units, allowing the computation of
unbiased estimators [70].
To ensure that sampled units were distributed across the main biogeographic regions,
a first level of stratification was applied based on the biomes defined by the ecoregions
2017 map [71]. The original 14 biomes were aggregated into 8 major biomes: boreal
forests, deserts and xeric shrublands, Mediterranean, temperate forests, temperate savanna,
tropical forests, tropical savanna, and tundra (Figure 4). A second level of stratification was
applied based on the FireCCI51 BA extent for each calendar year. Thus, each TSA within
each biome and year was assigned to a high or low BA stratum following the approach
of [72]. Then, the sample’s allocation within
q the 16 resulting strata (8 biomes × 2 high-low strata)
was established proportionally to Nh BAh , where Nh is the number of units in stratum h
and BAh is the mean mapped BA for stratum h [73].
The reference data for the selected sites were obtained from pairs of Landsat-8 images
using a semi-automatic classification algorithm [74], followed by a visual inspection to
confirm the correct identification of burned patches. Consecutive pairs of images were
used to obtain long temporal reference data, covering several months, the entire fire season
 ing the approach of [72]. Then, the sample’s allocation within the 16 resulting strata (8
biomes × 2 high-low strata) was established proportionally to 𝑁 𝐵𝐴 , where 𝑁 is the
number of units in stratum h and 𝐵𝐴 is the mean mapped BA for stratum h [73].
The reference data for the selected sites were obtained from pairs of Landsat-8 im-
Remote Sens. 2021, 13, 4295 ages using a semi-automatic classification algorithm [74], followed by a visual inspection 10 of 25

to confirm the correct identification of burned patches. Consecutive pairs of images were
used to obtain long temporal reference data, covering several months, the entire fire
season or the year,
or the whole whole year, depending
depending on imageonavailability.
image availability. These
These long long temporal
temporal referencerefer-
data
ence data allow for an extended temporal overlap between reference
allow for an extended temporal overlap between reference and BA datasets, which and BA datasets,
makes
which makes
it possible to it possiblethe
improve to spatial
improve the spatial
accuracy accuracyof
assessment assessment of the BA
the BA products, products,
minimising
minimising
the impact ofthe
theimpact of the
product’s product’s
temporal temporal
reporting reporting
accuracy accuracy
in the spatial in the spatial
accuracy accu-
estimates.
racy estimates.
Further details Further details onmethods
on the validation the validation
can bemethods
found incan be found in [75].
[75].

Figure
Figure 4.
4. Biomes
Biomes representing
representing the
the first
first stratum
stratum of validation sampling.
of validation sampling. The
The black
black locations
locations show
show the
the 300
300 validation
validation sites
sites
that were selected for 2017, 2018, and 2019 (100 for each year).
that were selected for 2017, 2018, and 2019 (100 for each year).

Accuracy
Accuracy metrics
metrics were
were computed
computed based
based on on the
the error
error matrix,
matrix, which
which is widely used
by
by the
the scientific
scientific community
community for thematic accuracy assessment [76,77]. This This matrix
matrix was
derived
derived from the thecross
crosstabulation
tabulationofof reference
reference data
data andand
the the BA products.
BA products. The metrics
The metrics used
used to assess
to assess the product
the product accuracy
accuracy were were the coefficient
the Dice Dice coefficient
(DC) (DC) [78],
[78], the the relative
relative bias
bias (relB)
(relB)
and theand the commission
commission (Ce) and(Ce) and omission
omission (Oe)
(Oe) errors. DCerrors. DCthe
measures measures thebetween
similarity similarity
the
between the two sets of data: the higher its value, the more accurate the burned
two sets of data: the higher its value, the more accurate the burned category. relB informs category.
relB
aboutinforms
the biasabout theBA
of the bias of the BA
product product
relative relative
to the to the
reference reference
BA: BA:relB
a negative a negative
means relB
that
means
the BAthat the BA
product product underestimates
underestimates the BA compared
the BA compared to thedata,
to the reference reference
whiledata, while
a positive
relB shows an overestimation. Ce refers to the burned area mapped by the BA product but
not classified as burned in the reference data, while Oe indicates the burned area in the
reference data not mapped in the BA product. All the accuracy metrics and their associated
standard errors were estimated for the whole population, applying the formulas described
in [72,79].

2.7.2. Temporal Reporting Accuracy Assessment

A validation of the temporal reporting accuracy of the products was also performed,
aiming to measure how accurately the day-of-burn (JD in Table 2) was determined by the
BA product. A common approach is to use as reference the day-of-detection reported by the
active fires and compare with the day-of-detection reported by the corresponding pixel of
the target BA product [80,81]. However, this approach assumes independence between the
active fires and the burned area product. Therefore, since both the C3SBA10 and FireCCI51
products use MODIS active fires, the temporal reporting accuracy assessment was carried
out based on the active fires detected at the 375 m spatial resolution by the VIIRS sensor on
board Suomi-NPP satellite (VNP14IMGML C1 thermal anomalies product [82]). Similar to
the MODIS thermal anomalies (MCD14ML) dataset, the product includes the location of
Remote Sens. 2021, 13, 4295 11 of 25

the anomalies and a field to select the presumed vegetation fires. The temporal reporting
accuracy assessment was computed both globally and by biome.

2.8. Analysis of the Impact of the Temporal Resolution in Burned Area

Since the C3SBA10 product was derived from two different satellite combinations (S3A
for 2017 and 2018, and S3A and B for 2019, it was convenient to test the impact of different
image availability conditions on the product’s accuracy and consistency. To do this, the
results for 2019 of the BA algorithms were compared using as inputs only S3A or both S3A
and B. The 13 calibration tiles described in Section 2.3 were used for this comparison. For
this analysis, an additional sample of 43 Landsat TSAs was processed to obtain reference
fire perimeters to check the accuracy of the resulting products. Those perimeters were
obtained following the same methods as those used for the global validation, but in this
case only covering the calibration sites.

3. Results
3.1. Intercomparison between FireCCI51 and C3SBA10
3.1.1. Summary of the Annual Burned Area
Figure 5 shows the annual accumulated BA at 0.25◦ for both FireCCI51 (a) and
C3SBA10 (b) in the year 2019, and Figure 6 shows the latitudinal and seasonal contri-
bution of BA for the year 2019. The amount of BA detected per biome for the three
complete natural years can be found in Table 3.
The C3SBA10 product detected 3.77 × 106 , 3.68 × 106 , and 3.59 × 106 km2 of BA
for the years 2017, 2018 and 2019, respectively (Table 3). The FireCCI51 estimated more
BA in all years, with proportions ranging from 7.8% in 2019 to 17.3% in 2017. The main
contributor to global BA in both products was the tropical savanna, representing between
74 and 78% of global burns. Therefore, most global BA was located on a relatively small
latitudinal band between 20◦ N and 20◦ S (Figures 5 and 6) that covers part of South
America, Central America, Africa, South East Asia, Indonesia and the northern part of
Australia. Among these tropical regions, several studies have shown that the African
savannas are responsible for around 70% of global BA, followed by the Australian ones,
and the Brazilian Cerrado and the Orinoquia region in Colombia and Venezuela [9,83–86].
Almost all the disparities between the products in terms of total BA could be explained via
the differences found in this biome, where C3SBA10 detected 0.49 × 106 , 0.46 × 106 , and
0.26 × 106 km2 less BA, respectively, in the 3 years of the overlapping time series.
Tropical forests account for between 9 and 11% of total BA (average 0.35 × 106 km2
for C3SBA10 and 0.40 × 106 km2 for the FireCCI51, respectively). The influence of the
dry season in tropical biomes is quite evident in the products, with a marked fire period
from November to March in the Northern Hemisphere, and from June to October in the
Southern hemisphere (Figure 6). This trend slightly differs in South East Asia where the
fire season continues until May. A lesser contribution, but still important, comes from the
peatland areas of Kalimantan and Sumatra in Indonesia.
Another significant belt of fire activity can be found in the temperate forest and grass-
lands of the Northern Hemisphere (Figure 5), with relative maximum of around 50◦ N
(Figure 6). Both C3SBA10 and FireCCI51 estimated that, on average, the accumulated
contribution of these biomes to the global fire activity was around 7% (0.27 × 106 km2
and 0.29 × 106 km2 , respectively). Most of this BA comes from the grasslands on the
Asian steppe, which is dominated by large fires, although there is also important agricul-
tural activity [87,88]. The central plains of United States also exhibit fire activity linked to
croplands, as do the forest areas of the north-western territories. In the Southern Hemi-
sphere, the BA derived from the extreme wildfires of 2019–2020 in the temperate forests
of south-eastern Australia [8] is noticeable in Figure 5. The monthly variability in BA in
temperate areas of the Northern Hemisphere is marked by both agricultural practices and
dry summer conditions, leading to two separated fire activity periods: one in March–April
and the other in July–August.
 Remote Sens. 2021, 13, 4295 12 of 25
Remote Sens. 2021, 13, x FOR PEER REVIEW 12 of 26

Figure 5. FireCCI51
Figure (a) and
5. FireCCI51 C3SBA10
(a) and (b)(b)annual
C3SBA10 annualaccumulated BAfor
accumulated BA forthe
theyear
year 2019
2019 ◦ spatial
at 0.25°
at 0.25 spatial resolution.
resolution.

Table 3. Burned area per biome derived from the C3SBA10 product, and relative differences against the predecessor global
Tropical forests account for between 9 and 11% of total BA (average 0.35 × 106 km2
BA product FireCCI51.
for C3SBA10 and 0.40 × 10 km for the FireCCI51, respectively). The influence of the dry
6 2

season in tropical
C3SBA10 biomes
Burned is Area
quite(km
evident
2) in the products, with
Difference a marked fire period from
C3SBA10–FireCCI51
Biome 1 1 2 1 1
2017 2018 2019 2017
November to March in the Northern Hemisphere, and from June to October 2018 2019 2 in the
Tropical savanna Southern hemisphere2,801,290
2,782,564 (Figure 6). This trend slightly
2,701,210 −17.7% differs in−South
16.4% East Asia
−9.5%where the
Tropical forest 335,032
fire season continues until313,597 −22.3% but still−important,
411,926contribution,
May. A lesser 17.6% −5.2% from the
comes
Temperate savanna 210,382 135,370 164,995 −9.8% −3.9% −0.4%
peatland258,622
Deserts and xeric shrubland areas of Kalimantan
148,073 and Sumatra
113,707 in Indonesia.
−20.7% −19.4% −2.8%
Temperate forest Another
89,537 significant belt of fire
111,662 activity can−8.9%
110,884 be found in the temperate
−11.5% −1.0%forest and
Boreal forest grasslands 61,173
of the Northern74,014 Hemisphere 90,503 (Figure 5), +4.5% +2.4%
with relative maximum+4.4%
of around 50°
Mediterranean −18.9% −32.8% −16.8%
N (Figure27,421
6). Both C3SBA10 8884 24,975
and FireCCI51 estimated that, on average, the accumulated
Tundra 5734 1697 13,044 +3.8% +4.1% +13.1%
contribution of these biomes to the global fire activity was around 7% (0.27 × 106 km2 and
Global 3,770,465 3,594,588 3,631,243 −17.3% −15.7% −7.8%
0.29 ×1 106 km2, respectively). 2
Most of this BA comes from the grasslands on the Asian
Only S3A satellite available. Both S3A and B satellites available.
steppe, which is dominated by large fires, although there is also important agricultural
activity [87,88]. The central plains of United States also exhibit fire activity linked to
croplands, as do the forest areas of the north-western territories. In the Southern Hemi-
sphere, the BA derived from the extreme wildfires of 2019–2020 in the temperate forests
of south-eastern Australia [8] is noticeable in Figure 5. The monthly variability in BA in
temperate areas of the Northern Hemisphere is marked by both agricultural practices
Remote Sens. 2021, 13, 4295 13 of 25

The deserts and xeric shrublands’ amount of BA detected by C3SBA10 and FireCCI51
products showed a similar contribution to that of temperate areas, although with much
more inter-annual variability. The BA for 2017 was 0.26 × 106 and 0.31 × 106 km2 , respec-
tively, while for 2019 this number decreased by more than a half (0.11 × 106 km2 according
to both products) (Table 3). Two areas on the Earth are responsible for most of this BA:
central and southern Kazakhstan are affected by large fires in summer (July–August), while
the xeric shrublands and grasslands of central and western Australia show fire activity
from October to January.
In the highest latitudes (>60◦ N), the boreal forests showed an increase in BA for the
three-year period, oscillating from 61,173 and 58,533 km2 in 2017 to 90,503 and 86,711 km2
in 2019, according to C3SBA10 and FireCCI51, respectively, which means a 50% increase. In
the Tundra region, BA showed an even higher increase, as in 2019 the products estimated
7 times more BA than in the previous year (Table 3). In fact, in 2019, the contributions of
these two biomes were almost as much as that of temperate forests. Figure 6 shows how
the fire season in the northern latitudes has a clear seasonal pattern in both products, with
the highest activity in the summer months (July–August).
Finally, the BA in Mediterranean areas showed the least agreement between products,
with the relative differences ranging from −17% (2019) to −33% (2018). The larger differ-
ences in 2018 are in line with the significant contrast between the BA of that year and of
the other two years, which was noticeable in both products. This inter-annual variation
seemed to be related to the extreme fire seasons in Portugal in 2017 and in 2019.

3.1.2. Spatial and Temporal Agreement

Even though the annual BAs detected by the global BA products differed through the
study period (2017–2019), the correlation showed a significant agreement between them at
all comparison grid sizes (Table 4). For instance, the correlation at the finest comparison
resolution (0.05◦ ) was found to be 0.937 in the year with lowest agreement (2017), and 0.952
in the year with the highest agreement (2019). This agreement increased when the spatial
resolution of the comparison was decreased, reaching a 0.99 correlation for all years at the
0.5◦ grid size. However, the slope showed a clear bias towards FireCCI51, pointing to an
underestimation of the C3SBA10 product at the global and annual scales, confirming the
differences in the accuracy metrics.
There is a clear improvement in the agreement between products in 2019 as well.
The correlation metric was the highest in that year (from 0.95 at 0.05◦ to 0.99 at 0.50◦ grid
resolutions), as was the slope (around 0.92 for all grid resolutions), which meant an increase
in the amount of BA detected by the C3SBA10. The RMSE for that year was also the lowest,
indicating that the availability of more images from both S3s made the differences between
the BA products smaller.
Through the eight biomes, the lowest correlations at the annual scale were those of
deserts and xeric shrublands, with a 0.86 correlation at 0.05◦ grid resolution (Figure 7e).
Tropical savanna showed the highest annual agreement, with correlations from 0.96 (at 0.05◦ )
to 0.99 (at 0.50◦ ). However, it had the highest RMSE of the different biomes (2.28 km2 at 0.05◦ )
due to the large amount of BA per grid cell. Boreal forest and tundra showed very high
agreement as well, with correlation coefficients between 0.94 and 0.98 for all grid sizes and
the lowest RMSE, i.e., 0.27 km2 for boreal forest and 0.14 km2 for tundra at 0.05◦ . The rest
of the biomes presented correlations between 0.87 and 0.89 at the 0.05◦ spatial resolution
and above 0.90 from 0.10◦ upwards.
The four biomes that were unbiased or showed a little bias towards C3SBA10 were
those located at the highest latitudes (boreal and tundra) and in temperate areas, with
slope values near to 1. The rest of the biomes showed an underestimation of the C3SBA10,
although in different proportions. The most noticeable underestimation was linked to the
Mediterranean biome (Figure 7f), with a maximum slope of 0.86 at the 0.05◦ grid resolution.
The rest of biomes presented slopes ranging from 0.89 to 0.97.
mote Sens. Remote Sens. 2021, 13, 4295
2021, 13, x FOR PEER REVIEW 14 of 25
14 of

Figure 6. Monthly
Figureburned area
6. Monthly for 2019
burned distributed
area for along
2019 distributed latitudes
along determined
latitudes byFireCCI51
determined by FireCCI51(a) (a)
andand C3SBA10
C3SBA10 (b). (b).

Table 4. Global correlation coefficient between FireCCI51 and C3SBA10, slope and RMSE for the
There is a clear improvement in the agreement between products in 2019 as w
years 2017, 2018 and 2019 at 0.05◦ , 0.10◦ , 0.25◦ and 0.50◦ grid sizes.
The correlation metric was the highest in that year (from 0.95 at 0.05° to 0.99 at 0.50° g
resolutions), as was the Yearslope (around 0.92 for all◦ GridgridSize
resolutions), which meant an
0.05◦ 0.10 0.25◦ 0.50◦
crease in the amount of BA detected by the C3SBA10. The RMSE for that year was a
Pearson’s r 2017 1 0.937 0.962 0.981 0.988
the lowest, indicating2018 that 1 the availability
0.943
of more
0.966
images0.983from both S3s0.989
made the d
ferences between the 2019 BA 2products0.952 smaller. 0.972 0.986 0.992
Slope the eight 2017 1 0.870 0.864 0.858annual scale
0.856 were those
Through biomes, the lowest correlations at the
2018 1 0.882 0.875 0.870 0.868
deserts and xeric shrublands, 2019 2
with0.923
a 0.86 correlation0.919 at 0.05°
0.916grid resolution
0.915 (Figure 7
Tropical RMSE
savanna showed 2017 the highest
1 1.347 annual agreement,
4.029 with
17.810 correlations
57.404 from 0.96
2018 1 1.279 3.807 16.800 54.338 biomes (2
0.05°) to 0.99 (at 0.50°). However, 2
it had the highest RMSE of the different
2019 1.073 3.060 12.577 38.092
km2 at 0.05°) due to the large amount of BA per grid cell. Boreal forest and tund
1 Only S3A satellite available. 2 Both S3A and B satellites available.
showed very high agreement as well, with correlation coefficients between 0.94 and 0
for all grid sizes and the lowest RMSE, i.e., 0.27 km2 for boreal forest and 0.14 km2
tundra at 0.05°. The rest of the biomes presented correlations between 0.87 and 0.89 at t
0.05° spatial resolution and above 0.90 from 0.10° upwards.
Remote Sens. 2021, 13, 4295 15 of 25

At the monthly scale, the agreement between products showed noticeable changes.
The most stable biome throughout the year was tropical savanna (Figure 7b), since it was
the most frequently burned biome and it was where most of the global BA was located.
Actually, the lowest correlation was found in the months that contributed less to annual
BA, i.e., February, March, April and May, with correlations from 0.86 to 0.89 at the finest
comparison resolution, respectively. The rest of the months showed correlations above
0.92 at the 0.05◦ grid size, and were higher at the rest of the compared resolutions. In fact,
at 0.50◦ , the lowest correlation, which was found in April, showed a value higher than 0.95.
A systematic underestimation of the C3SBA10 was noticeable through all the grid cells that
belong to tropical savanna, with slopes between 0.90 (September and December) and 0.98
(October) at the 0.05◦ spatial resolution. The influence of tropical savanna was clear in the
global metrics, which can in some cases obscure subjacent trends in other biomes.
In most cases, the correlation increased with the proportional contribution to the
annual BA. This was the case of boreal forests (Figure 7g) and tundra (Figure 7h), whose
peak months clearly matched the summer months, i.e., July and August, which were
responsible for 70% and 77% of the annual BA in those biomes, respectively, and, at the
same time, showed the highest agreement (Pearson’s r > 0.91). However, not all biomes
had a marked unimodal fire season, since they encompassed different continents and
hemispheres, therefore indicating that fire activity is affected by diverse climatic and
human conditions. In the case of deserts and the xeric shrubland biome (Figure 7e),
two separate peaks were found, one in January (20% of annual BA, Pearson’s r = 0.83,
slope = 1.12 at 0.05◦ grid size) and another in July (19% of annual BA, Pearson’s r = 0.92,
slope = 0.95 at 0.05◦ ). Something similar occurred in tropical forests (Figure 7a) where
January (13% of annual BA, Pearson’s r = 0.87, slope = 0.87 at 0.05◦ ) and August (14% of
annual BA, Pearson’s r = 0.92, slope = 0.95 at 0.05◦ ) had the maximum monthly BA values.
The tropical savanna showed two clear fire seasons as well (Figure 7b), following the dry
seasons of each hemisphere.
The agreement shown between C3SBA10 and FireCCI51 was not as stable in the case
of temperate and Mediterranean biomes. In the former case, there was a clear discrepancy
in the monthly contribution of April to the annual BA. In the case of temperate forests
(Figure 7c), the monthly contribution corresponded to 21.3% for C3SBA10, while FireCCI51
estimated it at 16.8%. The same occurred in the temperate savanna (Figure 7d), where
the contribution increased from 20.4% (FireCCI51) to 26.8% (C3SBA10). However, this
discrepancy affected mainly the RMSE, which was much higher for this month than for the
rest of the months in both biomes (RMSE = 0.44 km2 in savanna and RMSE = 0.27 km2 in
forest at 0.05◦ grid size). In the Mediterranean biome (Figure 7f), from July to November,
there was a significant increase in the bias towards FireCCI51, losing the stability shown in
the first half of the year, and with slope values as low as 0.70, e.g., in the case of October.

3.2. Validation
3.2.1. Spatial Assessment
Table 5 shows the estimated accuracy metrics and their standard errors (SE) for the
validation dataset. The Dice coefficient varied from 61.7 in 2019 (SE = 2.9) to 64.8 (SE = 2.9)
in 2018 for the C3SBA10 product. These values were similar to those obtained for the
FireCCI51 product, although this showed accuracy values that were higher for the years
2017 and 2018, but not substantially so for 2019 (Table 5), when the two S3 satellites were
available. In general terms, C3SBA10 showed lower commission errors (Ce) and higher
omission errors (Oe) than FireCCI51 throughout the three-year period. Both products
showed negative relative bias for all the years, indicating a systematic underestimation of
BA, which is common to other global BA datasets [89].
Although three years are not enough to extract significant conclusions about the
temporal trends of accuracy, a similar behaviour can be observed in both products. Thus,
the year that presented the highest accuracy metrics in one product also did so in the
other product.
 Remote Sens. 2021, 13, x FOR PEER REVIEW 16 of 26

Remote Sens. 2021, 13, 4295 16 of 25

bility shown in the first half of the year, and with slope values as low as 0.70, e.g., in the
case of October.

Figure7.7.Monthly
Figure Monthlypercent
percentBA
BAcontribution,
contribution,Pearson’s
Pearson’scorrelation
correlationand
andbias
bias(slope)
(slope)between
betweenthe
theFireCCI51
FireCCI51and
andC3SBA10
C3SBA10
products per biome (sub-figures (a–h)) and grid size for the year 2019.
products per biome (sub-figures (a–h)) and grid size for the year 2019.

3.2. for
Table 5. Accuracy metrics Validation
the years 2017, 2018 and 2019. Standard errors are shown between parentheses.
3.2.1. Spatial Assessment
1
2017 2018 1 2019 2
Accuracy Metrics
Table 5 shows
FireCCI51 the estimated
C3SBA10 accuracy metrics
FireCCI51 and theirFireCCI51
C3SBA10 standard errorsC3SBA10
(SE) for the
Dice coefficient (DC) validation
66.9 (2.3) dataset. 62.3The
(2.6)Dice coefficient
69.2 (2.7) varied 64.8
from(2.9)
61.7 in 2019 (SE
63.9 = 2.9) to 64.8
(2.8) 61.7(SE
(2.9)= 2.9)
Commission error (Ce) in
21.42018
(2.2) for the 19.5
C3SBA10
(2.1) product. These values
15.7 (1.4) 13.1 were
(1.3) similar to(1.7)
20.8 those obtained for the
18.6 (1.7)
Omission error (Oe) FireCCI51
41.8 (3.1) product, although this
49.2 (3.2) 41.3showed
(3.4) accuracy values that
48.3 (3.4) were
46.5 (3.4)higher for
50.3the
(3.4)years
Relative bias (relB) −2017 and 2018,−but
26.0 (4.1) 36.9not
(4.0) −30.4 (3.3)
substantially so for 2019 −40.5 −32.5
(3.5)5), when
(Table the(3.4) −39.0 (3.5)
two S3 satellites were
available.
1 Only S3AIn general
satellite terms,
available. C3SBA10
2 Both S3A andshowed
B satelliteslower commission errors (Ce) and higher
available.
omission errors (Oe) than FireCCI51 throughout the three-year period. Both products
3.2.2.
showed Temporal
negative Reporting Accuracy
relative bias for allAssessment
the years, indicating a systematic underestimation of
BA, About
which is 19common
million active
to other fires wereBA
global used to estimate
datasets [89]. the temporal reporting accuracy
of theAlthough
three-yearthree period of both
years are products
not enough (Figure 8). Onsignificant
to extract average (2017–2019),
conclusionsthe results
about the
of C3SBA10
temporal and of
trends FireCCI51
accuracy,showed
a similarthat for deserts
behaviour can be and xeric shrubland,
observed tropicalThus,
in both products. and
temperate
the year thatsavanna and Mediterranean
presented biomes, around
the highest accuracy metrics90% of burned
in one product pixels
alsowere
did detected
so in the
within the first
other product. 10 days after the fire (Tables A1 and A2). This proportion was slightly lower
in temperate and tropical forests, where about 80% and 83% of the pixels were labelled
within
Table 5.10Accuracy
days of metrics
detection, respectively.
for the years 2017,The2018lessandaccurate biomes
2019. Standard wereare
errors theshown
boreal ones,
between
where only 69% (forest) and 68% (tundra) of the burned pixels were detected by C3SBA10
parentheses.
within the first 10-day period, and slightly less by FireCCI51.
Accuracy 2017 1 2018 1 2019 2
The analysis showed that, on average, the products were only able to detect more
thanMetrics
20% of the FireCCI51
cases within C3SBA10
0 or 1-day FireCCI51
difference inC3SBA10 FireCCI51
temperate savanna C3SBA10
and desert and
Dice coeffi-
xeric shrubland66.9 biomes. This accuracy substantially increased if the difference was set to a
(2.3) 62.3 (2.6) 69.2 (2.7) 64.8 (2.9) 63.9 (2.8) 61.7 (2.9)
cient (DC)
maximum of 3 days’ difference, whereat all biomes doubled the number of cases that were
Commission
detected within this time threshold.
21.4 (2.2) 19.5 (2.1) 15.7 (1.4) 13.1 (1.3) 20.8 (1.7) 18.6 (1.7)
error (Ce)
Omission 41.8 (3.1) 49.2 (3.2) 41.3 (3.4) 48.3 (3.4) 46.5 (3.4) 50.3 (3.4)
 boreal ones, where only 69% (forest) and 68% (tundra) of the burned pixels were detected
by C3SBA10 within the first 10-day period, and slightly less by FireCCI51
The analysis showed that, on average, the products were only able to detect more
than 20% of the cases within 0 or 1-day difference in temperate savanna and desert and
Remote Sens. 2021, 13, 4295 xeric shrubland biomes. This accuracy substantially increased if the difference was 17
setofto
25
a maximum of 3 days’ difference, whereat all biomes doubled the number of cases that
were detected within this time threshold.

Figure 8. Temporal reporting accuracy assessment of both FireCCI51 and C3SBA10 products. The box represents the
interquartile range (IQR), the lower and upper bounds being the 25th and 75th percentiles, respectively. The black line
within the box corresponds to the median value. The whiskers represent an additional area of 1.5 × IQR.

3.3. Impact of the Number of Input Images

As indicated, the assessment on the performance of the C3SBA10 related to the number
of S3 satellites available was carried out only on the 13 calibration tiles, using a dedicated
validation dataset. As expected, the availability of S3B images from 2019 had an important
influence on the results, both in terms of accuracy and BA estimations (Table A3). The
omission error for those 13 tiles decreased from 45.4% (vS3A) to 40.5% (C3SBA10), while
the Dice coefficient increased almost 3% (DC = 68.1 for C3SBA10). The relative bias was
also reduced from −32.9 (vS3A) to −25.1 (C3SBA10), guided by the mentioned reduction
in omission errors and a slight increase (below two percentage points) in the commission
errors. All this led to an increment in the estimation of BA in the 13 tiles from 509,166 to
551,782 km2 (+8.4%).
In line with these findings, the comparison performed in the previous sections showed
that the availability of both S3A and B, and hence a higher temporal resolution, had a
significant impact on the amount of BA detected by the C3SBA10, mainly in those biomes
where fire activity is frequent. This explains that the estimations between FireCCI51 and
C3SBA10 were much more similar in 2019 than in the other years. Actually, the differences
Remote Sens. 2021, 13, 4295 18 of 25

between the two products for 2019 were reduced by more than half in 2017 and 2018 for
tropical forests, temperate savanna, desert and xeric shrubland and temperate forests,
and by almost a half in tropical savanna (Table 3). In fact, if tropical savanna was not
considered, only 30,000 km2 of BA was underestimated by the C3SBA10 versus FireCCI51.
This underestimation was partially compensated for by the detection of more BA in boreal
forest and tundra (5305 km2 ). The availability of more images can also explain why the
difference in the omission error between the C3SBA10 and FireCCI51 was reduced from
more than 7% in 2017 and 2018 to below 4% in 2019. As a consequence, the difference
between the Dice coefficients in 2019 is 2.2%, half of that observed in the pre-S3B years. All
this led to the highest agreement in 2019, which showed the highest Pearson’s correlations,
the lowest bias towards FireCCI51, and the lowest RMSE.

4. Discussion
The transference of the FireCCI BA products to an operational service was challenging
as, when the implementation process began in 2018, they were still in development, and
had to be adapted to the new Copernicus missions. However, it is important to provide
a consistent BA product for the climate service because BA information is demanded by
atmospheric and carbon modellers. For this reason, it was decided to make a conservative
choice by transferring the experience from the latest version of FireCCI’s BA algorithms
(FireCCI51) to the C3S service, while adapting it to the new OLCI sensor on board the
S3 satellites. Since the FireCCI51 product was conceived as a research product, not an
operational one, it was processed only until 2019. Therefore, this option would guarantee
the continuity of the CCI BA time series from 2019 onwards. Thus, the main objective of
the C3SBA10 BA algorithm was to create a consistent product with FireCCI51.
The results of the intercomparison analysis prove this consistency between C3SBA10
and FireCCI51, with similar spatial and temporal trends. The annual BA data show a
correlation value of 0.95, with a minor underestimation (slope = 0.92 and RMSE = 1.07 km2 )
at 0.05◦ spatial resolution, the finest spatial resolution used in the comparison (Table 4).
The agreement between the products improved as the grid resolution for intercomparison
decreased (Table 4, and Figure 7). The main differences between the two products were
observed in the first two years of the time series (2017 and 2018) when only S3A was
operating, and therefore the temporal resolution of OLCI (2–3 days) was much lower than
the MODIS sensor (1 day). When the two S3 satellites were operating, in 2019, the BA
estimations between the two products were more similar, although C3SBA10 detected
0.28 × 106 km2 less BA than FireCCI51. Most of this difference (92%) was located in tropical
savanna, where a systematic underestimation of C3SBA10 was found. This tendency
towards underestimation was also observed in other global BA products (generally based
on coarse spatial resolution sensors) when comparing them with regional products based
on medium-resolution sensors. For instance, a continental BA product at 20 m derived
from Sentinel-2 (S2) for the year 2016 in Sub-Saharan Africa found that global BA products
significantly underestimate total BA, as they included 80% less burned area than the S2
BA product [90]. This was mainly caused by the poor detection of small fires (<100 ha) in
global products. Therefore, C3SBA10 is likely missing more small fires than FireCCI51 due
to its coarser spatial resolution (300 m vs. 250 m).
A deeper analysis of the year 2019 showed different trends in the consistency between
products among the biomes. It was found that the changes in the correlation between prod-
ucts were linked to the BA detected for each month and biome. Although this relationship
did not seem to be linear, i.e., a specific increase in BA did not mean a parallel increase in the
correlation, it was clear that the larger the BA detected, the higher the correlation between
products. The tropical savanna is an obvious example, and is where most of global BA was
found. The boreal regions are another example where the correlation was the highest in
the months with the largest BA, while it significantly decreased for the rest of the months.
The extreme wildfire event of Eastern Australia [8] clearly illustrated this phenomenon,
since it significantly increased BA and the agreement between products in December in the
Remote Sens. 2021, 13, 4295 19 of 25

Mediterranean biome, and in November and December in temperate forests. There were
some unusual cases as well, where higher BA did not necessarily mean higher correlation.
This was the case for the temperate areas, where the correlation decreased in April.
Temperate savanna and forest biomes cover almost all the Northern Hemisphere’s mid-
latitude croplands, located in the Russian Federation and Kazakhstan [87]. Cropland fires
are characterised as small and transient [91,92], which made moderate-resolution BA prod-
ucts (250–500 m) unable to properly characterise the extent of cropland BA patches [87,88].
Besides this, the human activity in those regions may create two separate peak months
(bimodal fire season), one of them taking place in sub-optimal weather conditions, and
thus at times when less valid images are available [93]. It is reasonable to assume that
the same issue was affecting both the FireCCI51 and C3SBA10 products, which clearly
increased the uncertainty in those regions and, hence, decreased the agreement. In fact, it
was very likely that the rest of the biomes were affected by a similar issue [94], manifesting
a decrease in the agreement of products from March to May (Figure 7) when agricultural
activity is high globally [92].
The validation exercise carried out through the overlapping three-year period (2017–2019)
showed that both global BA products presented higher omission than commission errors [75].
This trend agrees with the validation exercises of other global BA products [79,83,89], as
well as with the abovementioned inability of moderate-resolution BA products to detect
small fire patches. The omission errors presented in this paper were lower than those found
by previous authors for other global BA products, e.g., 72.6% for NASA’s standard product
in [89], and 81% and 71% for two consecutive versions of FireCCI products (FireCCI41
and FireCCI50) in [83]. However, it must be highlighted that these estimations are not
fully comparable to those presented in this paper, since all those studies used so-called
short units (reference data generated from a maximum period of 16 days) instead of long
units (reference data covering several months, as done in this study) to estimate the spatial
accuracy of the products [75]. Similarly to the amounts of BA detections, the availability of
the second S3 in 2019 had a direct impact on the accuracy of C3SBA10 (DC = 61.7 ± 2.9),
and hence the accuracy of this year was more similar to that estimated for FireCCI51
(DC = 63.9 ± 2.8).
As was mentioned at the beginning of the discussion, the main aim of the C3SBA10
was to produce, based on European satellites, a BA product that is consistent with its
precursor FireCCI51. However, the successful achievement of this requirement meant, at
the same time, that some limitations detected in the FireCCI products were maintained
in the new dataset. Among these limitations were the low temporal reporting accuracy
shown by FireCCI51 [30] and potential tiling effects. The former was indeed observed,
as only around 20% of the burned pixels were detected within ±1 day after the fire, 44%
within ±3 days and 87% within ±10 days. The reporting accuracy problems of C3SBA10
and FireCCI51 are related to the criteria used to create the monthly composites, which are
the starting point of both algorithms. To avoid the cloud and observation problems of the
daily images, monthly NIR composites were created by selecting the most suitable NIR
observation on a 20-day moving window centred around the date of the nearest active
fire. This moving window was only expanded when less than four cloud-free images were
found after the fire. Thus, in most cases, the search was limited to the first 10 post-fire days.
However, inside the searching window, the algorithm prioritises the separability of the
burned signal over the proximity of the observation to the date of the nearest active fire.
The most separable observation, although it theoretically should be, is not always found in
the day immediately after the fire due to angular effects. Although a temporal reporting
accuracy of ±10 days could be reasonable for some applications (e.g., dynamic vegetation
modelling), there are others, such as atmospheric emissions estimations, wherein the
precise date of burn is very critical. However, these applications commonly use grid BA
files, where BA is provided as the total per month in each grid cell. An additional post-
processing step could be applied, for instance, reallocating the BA estimations following
the active fires’ temporal distribution [9].
Remote Sens. 2021, 13, 4295 20 of 25

Regarding tiling effects, since both algorithms are processed following a standard tiling
system, border effects may appear between adjacent tiles. A recent study that analysed
this effect on the FireCCI51 and NASA’s standard BA products (MCD64A1 c6 [85]), which
follow the standard MODIS tiling system, showed that horizontal border effects were
noticeable in both products [95]. In the case of FireCCI51, border effects were mainly found
in three different areas, although the most substantial one was located in Northwest India,
near Pakistan. Although FireCCI51 presented a new methodology to estimate different
thresholds for each fire, the first step in the algorithm still uses a tile-based threshold to
perform the initial filtering of active fires. To do this, an unburned sample is selected from
those pixels located further than 10 km from the nearest active fire. In the Indian case, the
fire’s activity is concentrated in the southern part of the tile, and therefore, the unburned
sample is mainly composed of pixels falling in the Himalayan mountainous area. This
area is full of shadows, with very low NIR values, and hence the estimated threshold (10th
percentile of the sample) is too low, filtering almost all the active fires in that initial step.
Although C3SBA10 followed a different tiling system than the MODIS products, tiles are
equally distributed every 10◦ in latitude, and therefore the same horizontal border effects
appear in the C3SBA10 product as well.
The use of MODIS active fires may be seen as a drawback of C3SBA10, since the
MODIS sensors are expected to be decommissioned ca. 2023. MODIS active fires were
selected again for consistency reasons, as they were used by the FireCCI51 product for the
benefit of having a long time series [30]. The replacement of MODIS active fires with those
detected by other sensors will not become effective until the end (or severe degradation) of
the MODIS mission. The best options for replacement would be the VIIRS sensor, on board
Suomi-NPP and NOAA-20, or the Sea and Land Surface Temperature Radiometer (SLSTR)
on board S3. The VIIRS active fires application can considerably increase the detection of
small fires due to its improved spatial resolution of 375 m (versus 1 km of MODIS active
fires) [48,49], although the impact on the identification of burned patches should be studied,
due to the limitations of the moderate-resolution reflectance data in detecting small burned
patches. Currently, the lack of a morning overpass of the VIIRS sensor could also be a
limitation, particularly considering the diurnal cycle of tropical fires [96,97]. Using S3
SLSTR to obtain active fires would be greatly beneficial for the C3SBA10 product, as BA
processing would not require external sensors. Besides this, the current version of the
active fires algorithm seems well adapted to small fires [98]. However, this version of the
algorithm is night-time-only, and consequently, provides limited sampling of the actual
fire activity.

5. Conclusions
This paper describes the process that was followed to adapt the FireCCI51 global
BA product, which is based on MODIS surface reflectance, to the C3S service using S3
OLCI data. The resulting product, called C3SBA10, ensures the continuation of the CCI BA
component. A critical issue when generating long-term time series of ECVs is to ensure the
consistency between the products that may be derived from different sensors. From that
perspective, the inter-comparison with FireCCI51 showed the high agreement in both the
spatial and temporal trends of BA, as well as a similar accuracy, especially during 2019,
when both S3A and B were available. At the time of writing this paper, more than three
years (January 2017–2020) of global BA data were publicly available at the Climate Data
Store in pixel (300 m) and grid (0.25◦ ) format.

Author Contributions: Conceptualisation, J.L.-L. and E.C.; methodology, J.L.-L., M.F. and E.C.;
software, J.L.-L., G.K. and M.B.; validation, M.F. and J.L.-L., formal analysis, J.L.-L., investigation,
J.L.-L., data curation, J.L.-L., G.K., M.B. and M.F.; writing—original draft preparation, J.L.-L., M.L.P.
and E.C.; writing—review and editing, J.L.-L., M.L.P., M.F. and E.C.; supervision, E.C.; project
administration, M.L.P. and E.C.; funding acquisition, E.C. All authors have read and agreed to the
published version of the manuscript.
Remote Sens. 2021, 13, 4295 21 of 25

Funding: This study has been funded by the Service Contract for the provision of land CDRs and ICDRs of the
Copernicus Climate Change Service (contract no. ECMWF/COPERNICUS/2018/C3S_312b_Lot5_VITO/SC1),
the ESA Climate Change Initiative–Fire ECV (contract no. 4000126706/19/I-NB), and the Spanish
Ministry of Science, Innovation, and Universities through a FPU doctoral fellowship (FPU17/02438).
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: The C3SBA10 product presented in this paper includes two types of
BA files: monthly full-resolution continental tiles and monthly global grid files at an aggregated
resolution of 0.25 degrees, as described in Section 2.5 and Table 2. Both datasets, pixel and grid
data types, are freely available through the Copernicus Climate Data Store (CDS) repository (DOI:
https://doi.org/10.24381/cds.f333cf85, accessed on 21 October 2021 [99]), referenced as OLCI C3S
Burned Area v1.0. Users should note that FireCCI51 is also available through the same repository,
referenced as MODIS FireCCI v5.1.1cds.
Conflicts of Interest: The authors declare no conflict of interest. The funders had no role in the design
of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or
in the decision to publish the results.
Additionally: The reference files generated for the spatial global validation of the C3SBA10 product
are publicly available at https://doi.org/10.21950/BBQQU7 (accessed on 21 October 2021) [100].

Appendix A
Table A1. Average (2017–2019) temporal reporting accuracies of the C3SBA10.

Desert and
Reporting Tropical Tropical Temperate Temperate Boreal
Xeric Mediterranean Tundra Global
Accuracy Forest Savanna Forest Savanna Forest
Shrubland
0–1 days 16.9% 18.5% 14.0% 21.4% 24.9% 14.2% 6.3% 7.3% 17.4%
0–3 days 40.9% 48.2% 33.5% 49.7% 56.2% 37.6% 17.5% 19.3% 44.5%
0–5 days 59.5% 68.0% 51.0% 69.0% 75.0% 59.3% 32.0% 33.3% 63.6%
0–10 days 83.4% 89.1% 80.7% 93.2% 94.1% 91.1% 69.2% 68.0% 86.8%

Table A2. Average (2017–2019) temporal reporting accuracies of the FireCCI51.

Desert and
Reporting Tropical Tropical Temperate Temperate Boreal
Xeric Mediterranean Tundra Global
Accuracy Forest Savanna Forest Savanna Forest
Shrubland
0–1 days 16.2% 18.6% 9.7% 23.1% 31.0% 13.8% 5.2% 5.5% 17.8%
0–3 days 40.2% 48.7% 27.3% 51.6% 61.8% 35.7% 15.4% 14.7% 45.2%
0–5 days 59.3% 68.4% 46.2% 70.6% 78.6% 57.4% 29.4% 27.3% 64.4%
0–10 days 84.3% 89.4% 78.3% 93.0% 94.4% 92.3% 63.7% 62.1% 87.1%

Table A3. Accuracy metrics for 2019 of the 13 calibration tiles. C3SBA10 is the official C3S BA product,
which used both S3A and B images. Conversely, vS3A represents the version that uses only S3A.

Accuracy Metric vS3A C3SBA10

Dice coefficient 65.3 68.1
Commission error 18.7 20.5
Omission error 45.4 40.5
Relative bias −32.9 −25.1
Remote Sens. 2021, 13, 4295 22 of 25

References
1. Marlon, J.R.; Bartlein, P.J.; Daniau, A.-L.; Harrison, S.P.; Maezumi, S.Y.; Power, M.J.; Tinner, W.; Vanniére, B. Global biomass
burning: A synthesis and review of Holocene paleofire records and their controls. Quat. Sci. Rev. 2013, 65, 5–25. [CrossRef]
2. Turco, M.; Jerez, S.; Augusto, S.; Tarín-Carrasco, P.; Ratola, N.; Jiménez-Guerrero, P.; Trigo, R.M. Climate drivers of the 2017
devastating fires in Portugal. Sci. Rep. 2019, 9, 13886. [CrossRef]
3. Chuvieco, E.; Pettinari, M.L.; Koutsias, N.; Forkel, M.; Hantson, S.; Turco, M. Human and climate drivers of global biomass
burning variability. Sci. Total Environ. 2021, 779, 146361. [CrossRef] [PubMed]
4. Abatzoglou, J.T.; Williams, A.P.; Boschetti, L.; Zubkova, M.; Kolden, C.A. Global patterns of interannual climate–fire relationships.
Glob. Chang. Biol. 2018, 24, 5164–5175. [CrossRef]
5. Andela, N.; van der Werf, G.R. Recent trends in African fires driven by cropland expansion and El Niño to La Niña transition.
Nat. Clim. Chang. 2014, 4, 791–795. [CrossRef]
6. Daniau, A.-L.; Sánchez Goñi, M.F.; Martinez, P.; Urrego, D.H.; Bout-Roumazeilles, V.; Desprat, S.; Marlon, J.R. Orbital-scale
climate forcing of grassland burning in southern Africa. Proc. Natl. Acad. Sci. USA 2013, 110, 5069. [CrossRef] [PubMed]
7. Moritz, M.A.; Parisien, M.-A.; Batllori, E.; Krawchuk, M.A.; Van Dorn, J.; Ganz, D.J.; Hayhoe, K. Climate change and disruptions
to global fire activity. Ecosphere 2012, 3, 49. [CrossRef]
8. Bowman, D.; Williamson, G.; Yebra, M.; Lizundia-Loiola, J.; Pettinari, M.L.; Shah, S.; Bradstock, R.; Chuvieco, E. Wildfires:
Australia needs national monitoring agency. Nature 2020, 584, 188–191. [CrossRef] [PubMed]
9. van der Werf, G.R.; Randerson, J.T.; Giglio, L.; Van Leeuwen, T.T.; Chen, Y.; Rogers, B.M.; Mu, M.; Van Marle, M.J.E.; Morton, D.C.;
Collatz, G.J. Global fire emissions estimates during 1997–2016. Earth Syst. Sci. Data 2017, 9, 697–720. [CrossRef]
10. Ward, D.; Kloster, S.; Mahowald, N.; Rogers, B.; Randerson, J.; Hess, P. The changing radiative forcing of fires: Global model
estimates for past, present and future. Atmos. Chem. Phys. 2012, 12, 10857–10886. [CrossRef]
11. Yue, C.; Ciais, P.; Houghton, R.A.; Nassikas, A.A. Contribution of land use to the interannual variability of the land carbon cycle.
Nat. Commun. 2020, 11, 3170. [CrossRef]
12. Poulter, B.; Cadule, P.; Cheiney, A.; Ciais, P.; Hodson, E.; Peylin, P.; Plummer, S.; Spessa, A.; Saatchi, S.; Yue, C. Sensitivity of
global terrestrial carbon cycle dynamics to variability in satellite-observed burned area. Glob. Biogeochem. Cycles 2015, 29, 207–222.
[CrossRef]
13. Friedlingstein, P.; O’Sullivan, M.; Jones, M.W.; Andrew, R.M.; Hauck, J.; Olsen, A.; Peters, G.P.; Peters, W.; Pongratz, J.;
Sitch, S.; et al. Global Carbon Budget 2020. Earth Syst. Sci. Data 2020, 12, 3269–3340. [CrossRef]
14. Tosca, M.G.; Randerson, J.T.; Zender, C.S.; Flanner, M.G.; Rasch, P.J. Do biomass burning aerosols intensify drought in equatorial
Asia during El Niño? Atmos. Chem. Phys. 2010, 10, 3515–3528. [CrossRef]
15. Tosca, M.G.; Diner, D.J.; Garay, M.J.; Kalashnikova, O.V. Observational evidence of fire-driven reduction of cloud fraction in
tropical Africa. J. Geophys. Res. Atmos. 2014, 119, 8418–8432. [CrossRef]
16. Kelly, L.T.; Brotons, L. Using fire to promote biodiversity. Science 2017, 355, 1264. [CrossRef]
17. Enright, N.J.; Fontaine, J.B.; Bowman, D.M.J.S.; Bradstock, R.A.; Williams, R.J. Interval squeeze: Altered fire regimes and
demographic responses interact to threaten woody species persistence as climate changes. Front. Ecol. Environ. 2015, 13, 265–272.
[CrossRef]
18. Bowman, D.M.J.S.; Balch, J.K.; Artaxo, P.; Bond, W.J.; Carlson, J.M.; Cochrane, M.A.; D’Antonio, C.M.; DeFries, R.S.; Doyle, J.C.;
Harrison, S.P.; et al. Fire in the Earth system. Science 2009, 324, 481–484. [CrossRef] [PubMed]
19. Higuera, P.E.; Abatzoglou, J.T. Record-setting climate enabled the extraordinary 2020 fire season in the western United States.
Glob. Chang. Biol. 2021, 27, 1–2. [CrossRef]
20. Bojinski, S.; Verstraete, M.; Peterson, T.C.; Richter, C.; Simmons, A.; Zemp, M. The concept of essential climate variables in
support of climate research, applications, and policy. Bull. Am. Meteorol. Soc. 2014, 95, 1431–1443. [CrossRef]
21. Plummer, S.; Lecomte, P.; Doherty, M. The ESA Climate Change Initiative (CCI): A European contribution to the generation of the
Global Climate Observing System. Remote Sens. Environ. 2017, 203, 2–8. [CrossRef]
22. Chuvieco, E.; Mouillot, F.; van der Werf, G.R.; San Miguel, J.; Tanasse, M.; Koutsias, N.; García, M.; Yebra, M.; Padilla, M.;
Gitas, I.; et al. Historical background and current developments for mapping burned area from satellite Earth observation. Remote
Sens. Environ. 2019, 225, 45–64. [CrossRef]
23. Mouillot, F.; Schultz, M.G.; Yue, C.; Cadule, P.; Tansey, K.; Ciais, P.; Chuvieco, E. Ten years of global burned area products from
spaceborne remote sensing—A review: Analysis of user needs and recommendations for future developments. Int. J. Appl. Earth
Obs. Geoinf. 2014, 26, 64–79. [CrossRef]
24. Seiler, W.; Crutzen, P.J. Estimates of gross and net fluxes of carbon between the biosphere and the atmosphere from biomass
burning. Clim. Chang. 1980, 2, 207–247. [CrossRef]
25. Forkel, M.; Dorigo, W.; Lasslop, G.; Teubner, I.; Chuvieco, E.; Thonicke, K. A data-driven approach to identify controls on global
fire activity from satellite and climate observations (SOFIA V1). Geosci. Model Dev. 2017, 10, 4443. [CrossRef]
26. Forkel, M.; Andela, N.; Harrison, S.P.; Lasslop, G.; van Marle, M.; Chuvieco, E.; Dorigo, W.; Forrest, M.; Hantson, S.; Heil, A.; et al.
Emergent relationships with respect to burned area in global satellite observations and fire-enabled vegetation models. Biogeo-
sciences 2019, 16, 57–76. [CrossRef]
27. Hantson, S.; Arneth, A.; Harrison, S.P.; Kelley, D.I.; Prentice, I.C.; Rabin, S.S.; Archibald, S.; Mouillot, F.; Arnold, S.R.; Artaxo, P.
The status and challenge of global fire modelling. Biogeosciences 2016, 13, 3359–3375. [CrossRef]
Remote Sens. 2021, 13, 4295 23 of 25

28. Hantson, S.; Kelley, D.I.; Arneth, A.; Harrison, S.P.; Archibald, S.; Bachelet, D.; Forrest, M.; Hickler, T.; Lasslop, G.; Li, F.; et al.
Quantitative assessment of fire and vegetation properties in simulations with fire-enabled vegetation models from the Fire Model
Intercomparison Project. Geosci. Model Dev. 2020, 13, 3299–3318. [CrossRef]
29. Heil, A.; Pettinari, M.L. ESA Climate Change Initiative-Fire_cci D1.1 User Requirements Document (URD), Version 7.2. 2021.
Available online: https://climate.esa.int/media/documents/Fire_cci_D1.1_URD_v7.2.pdf (accessed on 21 October 2021).
30. Lizundia-Loiola, J.; Otón, G.; Ramo, R.; Chuvieco, E. A spatio-temporal active-fire clustering approach for global burned area
mapping at 250 m from MODIS data. Remote Sens. Environ. 2020, 236, 111493. [CrossRef]
31. Otón, G.; Lizundia-Loiola, J.; Pettinari, M.L.; Chuvieco, E. Development of a consistent global long-term burned area product
(1982–2018) based on AVHRR-LTDR data. Int. J. Appl. Earth Obs. Geoinf. 2021, 103, 102473. [CrossRef]
32. Thépaut, J.; Dee, D.; Engelen, R.; Pinty, B. The Copernicus Programme and its Climate Change Service. In Proceedings
of the IGARSS 2018-2018 IEEE International Geoscience and Remote Sensing Symposium, Valencia, Spain, 22–27 July 2018;
pp. 1591–1593.
33. Sayer, A.M.; Hsu, N.C.; Bettenhausen, C.; Ahmad, Z.; Holben, B.N.; Smirnov, A.; Thomas, G.E.; Zhang, J. SeaWiFS Ocean Aerosol
Retrieval (SOAR): Algorithm, validation, and comparison with other data sets. J. Geophys. Res. Atmos. 2012, 117. [CrossRef]
34. Hsu, N.C.; Jeong, M.J.; Bettenhausen, C.; Sayer, A.M.; Hansell, R.; Seftor, C.S.; Huang, J.; Tsay, S.C. Enhanced Deep Blue aerosol
retrieval algorithm: The second generation. J. Geophys. Res. Atmos. 2013, 118, 9296–9315. [CrossRef]
35. Sayer, A.M.; Hsu, N.C.; Lee, J.; Bettenhausen, C.; Kim, W.V.; Smirnov, A. Satellite Ocean Aerosol Retrieval (SOAR) Algorithm
Extension to S-NPP VIIRS as Part of the “Deep Blue” Aerosol Project. J. Geophys. Res. Atmos. 2018, 123, 380–400. [CrossRef]
[PubMed]
36. Hsu, N.C.; Lee, J.; Sayer, A.M.; Kim, W.; Bettenhausen, C.; Tsay, S.C. VIIRS Deep Blue Aerosol Products Over Land: Extending the
EOS Long-Term Aerosol Data Records. J. Geophys. Res. Atmos. 2019, 124, 4026–4053. [CrossRef]
37. Sayer, A.M.; Hsu, N.C.; Lee, J.; Carletta, N.; Chen, S.H.; Smirnov, A. Evaluation of NASA Deep Blue/SOAR aerosol retrieval
algorithms applied to AVHRR measurements. J. Geophys. Res. Atmos. 2017, 122, 9945–9967. [CrossRef] [PubMed]
38. Embury, O.; Good, S. Product User Guide and Specification–Sea Surface Temperature, D3.SST.1-v2.2_PUGS_of_v2SST_products_v6.0.
2020. Available online: https://datastore.copernicus-climate.eu/documents/satellite-sea-surface-temperature/v2.0/D3.SST.1-
v2.2_PUGS_of_v2SST_products_v6.0_APPROVED_Ver1.pdf (accessed on 21 October 2021).
39. Merchant, C.J.; Block, T.; Corlett, G.K.; Embury, O.; Mittaz, J.P.D.; Mollard, J.D.P. Harmonization of Space-Borne Infra-Red Sensors
Measuring Sea Surface Temperature. Remote. Sens. 2020, 12, 1048. [CrossRef]
40. Merchant, C.I. Sea Surface Temperature CCI Phase-II Algorithm Theoretical Basis Document (v2 Reprocessing), SST_CCI-ATBD-
UOR-203 (ATBD), Issue 3. 2019. Available online: https://climate.esa.int/media/documents/SST_cci_PUG_v2.pdf (accessed on
21 October 2021).
41. Merchant, C.J.; Embury, O.; Bulgin, C.E.; Block, T.; Corlett, G.K.; Fiedler, E.; Good, S.A.; Mittaz, J.; Rayner, N.A.; Berry, D.; et al.
Satellite-based time-series of sea-surface temperature since 1981 for climate applications. Sci. Data 2019, 6, 223. [CrossRef]
[PubMed]
42. Zhang, Z.; Zhao, L.; Lin, A. Evaluating the performance of Sentinel-3A OLCI land products for gross primary productivity
estimation using ameriflux data. Remote Sens. 2020, 12, 1927. [CrossRef]
43. Pastor-Guzman, J.; Brown, L.; Morris, H.; Bourg, L.; Goryl, P.; Dransfeld, S.; Dash, J. The Sentinel-3 OLCI Terrestrial Chlorophyll
Index (OTCI): Algorithm Improvements, Spatiotemporal Consistency and Continuity with the MERIS Archive. Remote Sens. 2020,
12, 2652. [CrossRef]
44. Alonso-Canas, I.; Chuvieco, E. Global Burned Area Mapping from ENVISAT-MERIS data. Remote Sens. Environ. 2015, 163,
140–152. [CrossRef]
45. López, G.; Muller, J.-P.; Potts, D.; Shane, N.; Kharbouche, S.; Fisher, D.; Lewis, P.; Brockmann, C.; Danne, O.; Krueger, O.; et al.
GlobAlbedo Algorithm Theoretical Basis Document Version 4.12 2013. Available online: http://globalbedo.org/docs/
GlobAlbedo_Albedo_ATBD_V4.12.pdf (accessed on 21 October 2021).
46. Giglio, L.; Schroeder, W.; Hall, J.V.; Justice, C.O. Modis Collection 6 Active Fire Product User’s Guide Revision B. 2018. Available
online: https://modis-fire.umd.edu/files/MODIS_C6_Fire_User_Guide_B.pdf (accessed on 21 October 2021).
47. Wolters, E.; Dierckx, W.; Lordache, M.D.; Swinnen, E. PROBA-V Products User Manual v. 3.01. 2018. Available online:
https://proba-v.vgt.vito.be/sites/proba-v.vgt.vito.be/files/products_user_manual.pdf (accessed on 21 October 2021).
48. Oliva, P.; Schroeder, W. Assessment of VIIRS 375m active fire detection product for direct burned area mapping. Remote Sens.
Environ. 2015, 160, 144–155. [CrossRef]
49. Schroeder, W.; Oliva, P.; Giglio, L.; Csiszar, I.A. The New VIIRS 375 m active fire detection data product: Algorithm description
and initial assessment. Remote Sens. Environ. 2014, 143, 85–96. [CrossRef]
50. Oliva, P.; Martin, P.; Chuvieco, E. Burned area mapping with MERIS post-fire image. Int. J. Remote Sens. 2011, 32, 4175–4201.
[CrossRef]
51. Merchant, C.J.; Paul, F.; Popp, T.; Ablain, M.; Bontemps, S.; Defourny, P.; Hollmann, R.; Lavergne, T.; Laeng, A.; de Leeuw, G.
Uncertainty information in climate data records from Earth observation. Earth Syst. Sci. Data 2017, 9, 511–527. [CrossRef]
52. Mittaz, J.; Merchant, C.J.; Woolliams, E.R. Applying principles of metrology to historical Earth observations from satellites.
Metrologia 2019, 56, 032002. [CrossRef]
Remote Sens. 2021, 13, 4295 24 of 25

53. Sayer, A.M.; Govaerts, Y.; Kolmonen, P.; Lipponen, A.; Luffarelli, M.; Mielonen, T.; Patadia, F.; Popp, T.; Povey, A.C.; Stebel, K. A
review and framework for the evaluation of pixel-level uncertainty estimates in satellite aerosol remote sensing. Atmos. Meas.
Tech. Discuss 2019, 13, 373–404. [CrossRef]
54. Chuvieco, E.; Pettinari, M.L.; Heil, A.; Storm, T. ESA CCI EVC Fire Disturbance: D1.2 Product Specification Document, Version 6.3.
2017. Available online: https://climate.esa.int/media/documents/Fire_cci_D1.2_PSD_v6.3.pdf (accessed on 21 October 2021).
55. Amos, J. Arctic Wildfires: How Bad Are They and What Caused Them? Available online: https://www.bbc.com/news/world-
europe-49125391 (accessed on 21 October 2021).
56. Evangeliou, N.; Kylling, A.; Eckhardt, S.; Myroniuk, V.; Stebel, K.; Paugam, R.; Zibtsev, S.; Stohl, A. Open fires in Greenland in
summer 2017: Transport, deposition and radiative effects of BC, OC and BrC emissions. Atmos. Chem. Phys. 2019, 2, 1393–1411.
[CrossRef]
57. Boschetti, L.; Eva, H.D.; Brivio, P.A.; Gregoire, J.M. Lessons to be learned from the comparison of three satellite-derived biomass
burning products. Geophys. Res. Lett. 2004, 31, L21501. [CrossRef]
58. Giglio, L.; Randerson, J.T.; van der Werf, G.R.; Kasibhatla, P.S.; Collatz, G.J.; Morton, D.C.; DeFries, R.S. Assessing variability and
long-term trends in burned area by merging multiple satellite fire products. Biogeo Sci. Discuss 2010, 7, 1171–1186. [CrossRef]
59. Humber, M.L.; Boschetti, L.; Giglio, L.; Justice, C.O. Spatial and temporal intercomparison of four global burned area products.
Int. J. Digit. Earth 2018, 12, 460–484. [CrossRef]
60. Turco, M.; Herrera, S.; Tourigny, E.; Chuvieco, E.; Provenzale, A. A comparison of remotely-sensed and inventory datasets for
burned area in Mediterranean Europe. Int. J. Appl. Earth Obs. Geoinf. 2019, 82, 101887. [CrossRef]
61. Bistinas, I.; Oom, D.; Sá, A.C.L.; Harrison, S.P.; Prentice, I.C.; Pereira, J.M.C. Relationships between human population density
and burned area at continental and global scales. PLoS ONE 2013, 8, e81188. [CrossRef] [PubMed]
62. Knorr, W.; Kaminski, T.; Arneth, A.; Weber, U. Impact of human population density on fire frequency at the global scale.
Biogeosciences 2014, 11, 1085–1102. [CrossRef]
63. Kaiser, J.; Heil, A.; Andreae, M.; Benedetti, A.; Chubarova, N.; Jones, L.; Morcrette, J.-J.; Razinger, M.; Schultz, M.; Suttie, M.
Biomass burning emissions estimated with a global fire assimilation system based on observed fire radiative power. Biogeosciences
2012, 9, 527–554. [CrossRef]
64. Kuenen, J.J.P.; Visschedijk, A.J.H.; Jozwicka, M.; Denier van der Gon, H.A.C. TNO-MACC_II emission inventory; a multi-year
(2003–2009) consistent high-resolution European emission inventory for air quality modelling. Atmos. Chem. Phys. 2014, 14,
10963–10976. [CrossRef]
65. Aleksankina, K.; Heal, M.R.; Dore, A.J.; Van Oijen, M.; Reis, S. Global sensitivity and uncertainty analysis of an atmospheric
chemistry transport model: The FRAME model (version 9.15.0) as a case study. Geosci. Model Dev. 2018, 11, 1653–1664. [CrossRef]
66. Hu, L.; Keller, C.A.; Long, M.S.; Sherwen, T.; Auer, B.; Da Silva, A.; Nielsen, J.E.; Pawson, S.; Thompson, M.A.; Trayanov, A.L.
Global simulation of tropospheric chemistry at 12.5 km resolution: Performance and evaluation of the GEOS-Chem chemical
module (v10-1) within the NASA GEOS Earth system model (GEOS-5 ESM). Geosci. Model Dev. 2018, 11, 4603–4620. [CrossRef]
67. Morisette, J.T.; Baret, F.; Liang, S. Special issue on global land product validation. IEEE Trans. Geosci. Remote Sens. 2006, 44,
1695–1697. [CrossRef]
68. Cohen, W.B.; Yang, Z.; Kennedy, R. Detecting trends in forest disturbance and recovery using yearly Landsat time series: 2.
TimeSync—Tools for calibration and validation. Remote Sens. Environ. 2010, 114, 2911–2924. [CrossRef]
69. Kennedy, R.E.; Yang, Z.; Cohen, W.B. Detecting trends in forest disturbance and recovery using yearly Landsat time series: 1.
LandTrendr—Temporal segmentation algorithms. Remote Sens. Environ. 2010, 114, 2897–2910. [CrossRef]
70. Gallego, F.J. Stratified sampling of satellite images with a systematic grid of points. ISPRS J. Photogramm. Remote Sens. 2005, 59,
369–376. [CrossRef]
71. Dinerstein, E.; Olson, D.; Joshi, A.; Vynne, C.; Burgess, N.D.; Wikramanayake, E.; Hahn, N.; Palminteri, S.; Hedao, P.; Noss, R.; et al.
An Ecoregion-Based Approach to Protecting Half the Terrestrial Realm. BioScience 2017, 67, 534–545. [CrossRef] [PubMed]
72. Padilla, M.; Stehman, S.V.; Chuvieco, E. Validation of the 2008 MODIS-MCD45 global burned area product using stratified
random sampling. Remote Sens. Environ. 2014, 144, 187–196. [CrossRef]
73. Padilla, M.; Olofsson, P.; Stehman, S.V.; Tansey, K.; Chuvieco, E. Stratification and sample allocation for reference burned area
data. Remote Sens. Environ. 2017, 203, 240–255. [CrossRef]
74. Bastarrika, A.; Alvarado, M.; Artano, K.; Martinez, M.; Mesanza, A.; Torre, L.; Ramo, R.; Chuvieco, E. BAMS: A Tool for
Supervised Burned Area Mapping Using Landsat Data. Remote Sens. 2014, 6, 12360–12380. [CrossRef]
75. Franquesa, M.; Lizundia-Loiola, J.; Stehman, S.V.; Chuvieco, E. Using long temporal reference units to assess the spatial accuracy
of global satellite-derived burned area products. Remote Sens. Environ. 2021. in review.
76. Congalton, R.G. A review of assessing the accuracy of classifications of remotely sensed data. Remote Sens. Environ. 1991, 37,
35–46. [CrossRef]
77. Stehman, S.V.; Foody, G.M. Key issues in rigorous accuracy assessment of land cover products. Remote Sens. Environ. 2019,
231, 111199. [CrossRef]
78. Dice, L.R. Measures of the Amount of Ecologic Association Between Species. Ecology 1945, 26, 297–302. [CrossRef]
79. Padilla, M.; Stehman, S.V.; Hantson, S.; Oliva, P.; Alonso-Canas, I.; Bradley, A.; Tansey, K.; Mota, B.; Pereira, J.M.; Chuvieco, E.
Comparing the Accuracies of Remote Sensing Global Burned Area Products using Stratified Random Sampling and Estimation.
Remote Sens. Environ. 2015, 160, 114–121. [CrossRef]
Remote Sens. 2021, 13, 4295 25 of 25

80. Boschetti, L.; Roy, D.P.; Justice, C.O.; Giglio, L. Global assessment of the temporal reporting accuracy and precision of the MODIS
burned area product. Int. J. Wildland Fire 2010, 19, 705–709. [CrossRef]
81. Campagnolo, M.L.; Libonati, R.; Rodrigues, J.A.; Pereira, J.M.C. A comprehensive characterization of MODIS daily burned area
mapping accuracy across fire sizes in tropical savannas. Remote Sens. Environ. 2021, 252, 112115. [CrossRef]
82. Schroeder, W.; Giglio, L. Visible Infrared Imaging Radiometer Suite (VIIRS) 375 m & 750 m Active Fire Products, Product User’s
Guide Version 1.4. 2018. Available online: https://viirsland.gsfc.nasa.gov/PDF/VIIRS_activefire_User_Guide.pdf (accessed on
21 October 2021).
83. Chuvieco, E.; Lizundia-Loiola, J.; Pettinari, M.L.; Ramo, R.; Padilla, M.; Tansey, K.; Mouillot, F.; Laurent, P.; Storm, T.; Heil, A.
Generation and analysis of a new global burned area product based on MODIS 250 m reflectance bands and thermal anomalies.
Earth Syst. Sci. Data 2018, 10, 2015–2031. [CrossRef]
84. Chuvieco, E.; Yue, C.; Heil, A.; Mouillot, F.; Alonso-Canas, I.; Padilla, M.; Pereira, J.M.; Oom, D.; Tansey, K. A new global burned
area product for climate assessment of fire impacts. Glob. Ecol. Biogeogr. 2016, 25, 619–629. [CrossRef]
85. Giglio, L.; Boschetti, L.; Roy, D.P.; Humber, M.L.; Justice, C.O. The Collection 6 MODIS burned area mapping algorithm and
product. Remote Sens. Environ. 2018, 217, 72–85. [CrossRef] [PubMed]
86. Roy, D.P.; Boschetti, L.; Justice, C.O. The collection 5 MODIS burned area product—Global evaluation by comparison with the
MODIS active fire product. Remote Sens. Environ. 2008, 112, 3690–3707. [CrossRef]
87. Hall, J.V.; Loboda, T.V.; Giglio, L.; McCarty, G.W. A MODIS-based burned area assessment for Russian croplands: Mapping
requirements and challenges. Remote Sens. Environ. 2016, 184, 506–521. [CrossRef]
88. Zhu, C.; Kobayashi, H.; Kanaya, Y.; Saito, M. Size-dependent validation of MODIS MCD64A1 burned area over six vegetation
types in boreal Eurasia: Large underestimation in croplands. Sci. Rep. 2017, 7, 4181. [CrossRef] [PubMed]
89. Boschetti, L.; Roy, D.P.; Giglio, L.; Huang, H.; Zubkova, M.; Humber, M.L. Global validation of the collection 6 MODIS burned
area product. Remote Sens. Environ. 2019, 235, 111490. [CrossRef] [PubMed]
90. Ramo, R.; Roteta, E.; Bistinas, I.; van Wees, D.; Bastarrika, A.; Chuvieco, E.; van der Werf, G.R. African burned area and fire
carbon emissions are strongly impacted by small fires undetected by coarse resolution satellite data. Proc. Natl. Acad. Sci. USA
2021, 118, e2011160118. [CrossRef] [PubMed]
91. McCarty, J.L.; Korontzi, S.; Justice, C.O.; Loboda, T. The spatial and temporal distribution of crop residue burning in the
contiguous United States. Sci. Total Environ. 2009, 407, 5701–5712. [CrossRef]
92. Randerson, J.; Chen, Y.; Werf, G.; Rogers, B.; Morton, D. Global burned area and biomass burning emissions from small fires. J.
Geophys. Res. Biogeosci. (2005–2012) 2012, 117, G040012. [CrossRef]
93. Benali, A.; Mota, B.; Carvalhais, N.; Oom, D.; Miller, L.M.; Campagnolo, M.L.; Pereira, J.M.C. Bimodal fire regimes unveil a
global-scale anthropogenic fingerprint. Glob. Ecol. Biogeogr. 2017, 26, 799–811. [CrossRef]
94. Vadrevu, K.P.; Lasko, K.; Giglio, L.; Schroeder, W.; Biswas, S.; Justice, C. Trends in Vegetation fires in South and Southeast Asian
Countries. Sci. Rep. 2019, 9, 7422. [CrossRef] [PubMed]
95. Liu, T.; Crowley, M.A. Detection and impacts of tiling artifacts in MODIS burned area classification. IOP SciNotes 2021, 2, 014003.
[CrossRef]
96. Giglio, L.; Kendall, J.D.; Mack, R. A multi-year active fire dataset for the tropics derived from the TRMM VIRS. Int. J. Remote Sens.
2003, 24, 4505–4525. [CrossRef]
97. Campagnolo, M.L.; Oom, D.; Padilla, M.; Pereira, J.M.C. A patch-based algorithm for global and daily burned area mapping.
Remote Sens. Environ. 2019, 232, 111288. [CrossRef]
98. Xu, W.; Wooster, M.J.; He, J.; Zhang, T. First study of Sentinel-3 SLSTR active fire detection and FRP retrieval: Night-time
algorithm enhancements and global intercomparison to MODIS and VIIRS AF products. Remote Sens. Environ. 2020, 248, 111947.
[CrossRef]
99. Lizundia-Loiola, J.; Franquesa, M.; Boettcher, M.; Kirches, G.; Pettinari, M.L.; Chuvieco, E. C3SBA10: OLCI C3S Burned Area v1.0.
Clim. Data Store 2020. [CrossRef]
100. Franquesa, M.; Vanderhoof, M.K.; Stavrakoudis, D.; Gitas, I.; Roteta, E.; Padilla, M.; Chuvieco, E. BARD: A global and regional
validation burned area database, V5. e-cienciaDatos 2020. [CrossRef]